Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2013

Replying to @pablochacin
yeah, it's unusually warm and sunny. Forecast for tomorrow: 14°C.
Replying to @sscdotopen
mmmh Bayesian methods also come with their very own challenges at scaling ;)
Replying to @ChrisDiehl
Ah well, all will be good. And the need to frameworkize will vanish. ;)
Replying to @ChrisDiehl
that I now have to get up at 6am workdays to bring my daughter to school also proved just as painful as I expected...
Replying to @ChrisDiehl
no, everything's fine. Holidays lined up favorably so I could take the next two weeks of to recharge.
Also, this didn't happen before the update with Android 4.1. OK let's just file a bug report. Oh wait... #wecant
Ok, is Android 4.3 more permissive to background tasks? Happened several times already, apps sitting in the back sucking the battery dry.
Replying to @lojikil
well, I admit being a bit more heavy on the social network side. But it worked before... Why, Samsung? Why, Google??
Replying to @lojikil
luckily i don't have those. But the growing number of Google Play X apps... . What is Google Play Kiosk anyway? ;)
Replying to @lojikil
thanks ;) I hate them because now I have to go into that mindset where you're always checking after your apps to find the bad one.
Replying to @eoinhurrell
@UltimateHurl I was only being sarcastic ;) but it does annoy me when you have to clear notifications x times (with x > 1).
.@muratk3n I've heard that one before. But Google is a company and you they don't have to publish results. They have to make money after all
Replying to @eoinhurrell
@UltimateHurl even more so when the state of read-ness is not synchronized properly!
I'm almost to the point where I want to dedicate my life to a consistent distributed notification framework for social apps.
Have an awful earworm which is "Let it snow" crossfading into "Rudolph the red nosed reindeer".
Replying to @markusandrezak
yeah, same here, my kids can't read (or know English), my wife is a anti-smartphone ;)
Just thumbed through "RESTful Java with JAX-RS 2.0". Looks like the kind of book where you find valuable info even if you have experience.
1
Replying to @superglaze
this is truly the stuff of nightmares. At least they have transparent processes if you're blocked accidentally. Oh wait...
The weirdest part of doing overdubs is to count you in to yourself. It's like little notes to self. "alright starting in 1.. 2.. 3.. 4.."
Replying to @ChrisDiehl
@ogrisel and then to fix all the numerical problems which the paper glossed over ;)
Replying to @ChrisDiehl
@ogrisel in a way it's a very simple idea. I think the key is to be able to interpret the score in a way which makes sense.
I suspect that most of those "the new firmware drains my battery in three hours" reports are just people exploring new features. For hours.
1
Is there a thing like the Gartner cycle for frameworks? Hadoop is definitely in the "so big everyone flocks to it" phase.
Lol set up Google Now which pops up notifications every fifteen minutes on how long it would take to go back home. #tryingtoworkmkay
1
Other new stuff: phone can now switch from wireless to 3G if wifi is slower than mobile Internet.
Yay, finally firmware upgrade from 4.1. (Jelly Bean) it 4.3 (still Jelly Bean). First impressions: the font is so sharp it hurts.
RT @PFCdgayo: Dear follower not really interested in academic fight clubs: Sorry. But if you are interested, the whole story: https://t.co/…
Replying to @ChrisDiehl
if the learner doesn't output class posteriori, a simple kernel density estimator is fitted first. A bit awkward maybe.
Replying to @ChrisDiehl
I know some papers which look at gradients on P(y|x) to see which is locally the most influential feature.
Replying to @iskander
uniformed guards? Oh man... I wondered how he got there. Possibly with his private helicopter ;)
Replying to @josephreisinger
@samkaufman Btw that there was a lot of talking about money instead of science is a complaint I heard from others as well.
RT @slyphon: If it walks like a duck, and quacks like a duck, it’s probably gonna throw exceptions at runtime.
Replying to @ChrisDiehl
I got 99% accuracy! Oh wait there's a bug, I'm always returning the same number...
RT @dataScienceRet: R: the good parts - Douglas Crockford’s JavaScript: the Good Parts book is now a classic. It’s also a tiny,... http://t…
Replying to @AnnaKendrick47
I always wondered how his relationship with Bella Swan in Adventureland led to her dating a vampire.
1
RT @AnnaKendrick47: Zuckerberg kept pestering me to add jokes to our presentation. So I actually said "I love you in Zombieland." He laughe…
Sometimes there's a new feature somewhere and I'm just like what do you WANT from me?? #InstagramDirect
1
On my way to the Microsoft Accelerator in Berlin to talk about ML. Let's see how that pans out.
1
@albert_swart methinks the problem with these numerics based algorithms is that you always get an answer. But does it make sense?
@albert_swart I'll be giving a talk tonight at the Berlin Microsoft Accelerator on ML. I'll make sure to include this ;)
Sometimes I think at least 30% of any data related work consists of asking myself "do these results make sense or is there still a bug?"
1 1
Replying to @doomsuckle
just did a two pages per sec scan. Will report on levels of narcissistic ramblings later ;)
Replying to @huitseeker
Hehe, yes. But given the late flood of data science books I was a bit skeptical how much originality to expect.
Replying to @roidrage
definitely ;) Some focus more on high level or business stuff, others on specific proglangs like Python, etc.
Interestingly, not all books with Data Science in the title are equal. Need to write a review when I have time.
1
RT @glouppe: Just presented "Scikit-Learn: Machine Learning in the Python ecosystem" at MLOSS #NIPS2013 Find the notebook at http://t.co/SD…
Replying to @alansaid
@twitter whoa, I know what you mean. This is the iOSification of Android...
It's that time of the year again where you optimize GC performance looking at jvisualvm graphs all afternoon long. <3
1
Still can't quite fathom that Zuckerberg himself showed up at the #NIPS2013 conference. It's really a story of obscurity to max relevance.
1 1
The thing that bugs me about the chromebook is that apart from the price it only takes features away from notebooks.
I think I'll start considering being able to give arbitrary length talks on any given subject an essential skill.
Today in a meeting five senior researchers were asked to give a five minute overview of their project. No one took less than fifteen minutes
1 3
RT @ylecun: Big News: I am joining Facebook as director of the new AI laboratory. I will still be Professor at NYU part time. http://t.co/m…
@gappy3000 @ChrisDiehl oh, oh, I just had an idea for an even better paper: "On Bayesian decision theory and the axiom of choice"
Played UNO the card game with my kids. My 3yo was very upset when he got rid of all his cards. We: "you won!" He: "no I want more caaards!"
3
That moment when you're the only person in the subway wearing something which is neither black nor grey.
1
@bastianventhur I can show it. It was quite funny because it was so extreme. Guess we were lucky, though.
@bastianventhur you mean because of the "TU is on shutdown. We've retracted into the bunker" email yesterday? ;)
Whoa, lightning in #berlin. At 2°C. Maybe there's more energy in that storm than I thought...
With all the rekindled interest in alternative programming languages, it's probably time to tell the world about Rhabarber.
2
Started writing a blog post on why Twitter isn't that big in Germany. But tapped on some deeper issues in my German soul... Have to ponder..
RT @ChrisDiehl: Using graph degeneracy to spot collaboration groups in the enterprise - My final blogpost on this thread - http://t.co/cGIt…
Windows in a full screen VM on Linux is the new... I don't know what it is, but it's awesome!
1
Replying to @dominik
mehr kann man nicht verlangen. Ich komme ja eigentlich als Köln. Ob es langsam Zeit ist, meine life decisions zu überdenken?
Replying to @dominik
bei dem angekündigten Unwetter frag ich mich eh ob das überhaupt noch weitergeht. Mit allem und so.
Replying to @danielerasmus
Hm. Either you're really trying to loose weight or your account has been hacked...
Replying to @superglaze
yeah I see. Embarrassment is probably the key word here. You don't want to be caught using Twitter in Germany (yet).
Replying to @superglaze
although I can't really say why Germans don't like Twitter. Somehow it's not considered to be "serious" enough.
1
Replying to @superglaze
Somehow this reminds me that most people working for Twitter Germany are PR/Media people.
Just realized I had put bacon on the shopping list last Saturday. I had bacon in the fridge for four days and completely forgot about it!!!
The What's New page for IDEA 13 reminds me very much of ads for detergents. "Now even whiter than white!" ;)
1
Replying to @ChengSoonOng
maybe we should make it more known we're sitting on terabytes of Twitter data.
IIRC, tracking a keyword with Topsy costs ¢20 per day, so $200m would be enough to track about 28,000 keywords for 100 years.
1
Replying to @fhuszar
seems like a lot of money. But then what does that mean to Topsy own service. I guess we'll see...
Replying to @fhuszar
I'm just wondering whether they are really after the real time technology for analysing their own data. $200m for Twitter analytics
Replying to @fhuszar
.@fhuszar keep wondering, though, whether Topsy was probably having difficulties selling their service at the price they wanted.
Replying to @fhuszar
.@fhuszar Hehe. How about at redoing that with Tim Cook with a chef's head photoshopped and saying "because I'm the cook that's why!" ;)
And suddenly, without any build up or justification whatsoever DJ Random decides to play Paradise City. \m/ (>. <)
1
RT @wohali: ENTERPRISE, FUCK YEAH™ : "JSONx is an IBM® standard format to represent JSON as XML." I can't make this shit up, folks.
Someone just asked me if I really were a frequentist. And born after 1950. Now *that's* a discussion I haven't had yet ;)
In terms of new data necessary to change ones mind, being opinionated may just be as bad as being inexperienced.
I can at least say from personal experience that reading the American's drunk late night tweets in the morning can be oddly funny at times.
I wonder if all those American people reading us tired European') s tweets in the morning has any effect on their mood.
1
That feeling when you stand up from your chair with your headphones in and there's a slight electrostatic discharge through your ears.
Gotta say, the Twitter app's feature to have per-user notifications is pretty nifty. #readingyourtweets
I *really* need to set the latency on my mental autocomplete to more the 100ms it is right now apparently.
Replying to @pavlobaron
if it comes from the trenches, no objections. If upper marketing decides to roll out XYZ because it's hip, we have a problem ;)
Replying to @pavlobaron
pragmatic more in the sense of "no we don't need the latest clojure/Julia/python based framework, I can just do it with R"
Dude, couldn't resist the Cyber Monday sale on O'Reilly. Restocked on Data Science and Android books ;)
I have to confess my mental model of Facebook's offerings is sketchy at best. #timelinesright
Replying to @wortstock
@simon_lorenz ihr habt in Köln ja noch mal so 30 Minuten mehr Sonnenlicht, hier in Berlin ist's um vier schon richtig dunkel :(
WWtND: what would the NSA do? What you should ask yourself when thinking about potential security holes in your software.
Replying to @superglaze
@gigastacey my mental autocomplete said "how the Internet of things could kill humanity" ;)
RT @NeinQuarterly: Until there is truth. Until there is beauty. There is, at least, German grammar.
Shaving will be the first thing to do when I get home this evening. Not that I have any beard to speak of. But it's killing me.
Replying to @grok_
@thinkberg just go with the net value of the paper. Did the same when customs wanted to know the value of conference proceedings.
I still remember people making fun of me and my "manager phone" when I had a Nokia E61i. Now practically everyone has a smartphone.
@thinkberg every time I hear the word "onboarding", there is a weak association with the word "waterboarding" :(
1 2
The best minds of our generation are writing texts to get more funding. I'm all for paper trail, but there has got to be a better way.
1 1
Replying to @karpathy
I have to get up early now because my kids have started to go to school, but I just can't get used to it :(
Replying to @karpathy
halfway through your tweet I was still thinking you were talking about getting up early to code ;)
@albert_swart everything's fine, I just mistook n as size of universe where it was just the number of elements to store.
Well, and since ln 2 about 0.69, no of bins should be at least 2.88 more than no of elements in universe so that two hash fcts is optimal.
1
I'm not making this up, papers say the optimal number of bins k = ln 2 * #bits / #elements in universe.
1
Wait, wait. When using bloom filters, you should actually have more bits then distinct elements you are trying to hash???
@mdreid ironically, I'm not that sure whether that sentence was grammatically correct at all ;)
Kids, if you read this, producing coherent texts of any length on any subject is a super important job skill!!!
4 1
Sometimes I wished my teachers would have stressed more how important writing skills are for the real world - oh wait, they have!
4 2
Replying to @purbon
some may have given up, but there is no getting used to the dimness if you ask me ;)
Replying to @purbon
the lack of sun in the Berlin winter is definitely worse than the cold.
Replying to @mdreid
looks like birds weren't so dumb after all. It's easy if you can just fly somewhere warm yourself, though.
Replying to @octonion
@davidandrzej I fully agree. No single tool fits it all right now. Probably never will.
RT @octonion: @mikiobraun @xamat If you don't know and use at least Python and/or Ruby, R and SQL, you're not a real data scientist.
Replying to @cartazio
need to check back on the discussion on HN in an hour or so to see how far the discussion has escalated ;)
Replying to @cartazio
yep. Believe me I know. I ended up writing jblas, so actually it's more like JRuby+Java+Fortran. :(
Replying to @cartazio
Very true. Or at least being able to plug in faster code without much pain. Which was why I was considering JRuby+Java for a time.
Replying to @cartazio
ah, I should have said "in terms of language for data analysis". No need to trigger python lovers that easily ;)
Replying to @cartazio
I mean what has python really going in terms of language? Operator overloading?
Replying to @cartazio
if python told us one thing it's that the infrastructure of good toolkits is as important as the language.
Replying to @cartazio
R. Back then when we discussed alternatives to Matlab R never came up. I guess people wanted a full lang instead.
Replying to @cartazio
they certainly got that platform thing right for statisticians. Somehow, we never considered it...
Replying to @cartazio
I do. ;) thanks for standing up for me ;) Didn't talk about R at all. Ah, well...
RT @MrPhilHarrison: Congratulations to Dan Livingstone of NZ - you're officially the first Xbox One owner on the planet #XboxOneLaunch http…
@haiqus I thought the argument was that in SJSU when people have a diverse level of prior knowledge, MOOCs have a hard time suceeding.
@haiqus you read it that way? I'm probably too biased against MOOCs to blame the SJSU students.
Replying to @MaineC
yep, although the recompiling becomes slower and slower. And the templating (lucid?) is a bit strange. But static files are great!
Replying to @davidandrzej
hardly ideal. But in some areas (medicine?) the standard is something proprietary.
Replying to @davidandrzej
agreed. We had a long discussion about this at JMLR. In the end we said paid software is OK if the whole community is using it
Replying to @cartazio
hehe ;) but I fear that with age I became a post-languagist pragmatist.
But funny to think that I co-organized one of the first meetings in the ML community to discuss how to move beyond matlab.
Replying to @MassimoMorelli
and I always felt that Ruby was more expressive on the command line. But already back then Python had more libs like numpy.
Replying to @MassimoMorelli
at that time I was toying with a new language with configurable parser to get true matrix literals etc.
Someday I need to blog about how I tried to steer the ML community away from Python. Not that I had a lot of a chance. ;)
2
Being able to argue at a high level well really seems to be something you can only learn through experience.
Practice session for the #NIPS2013 poster spotlight of one of our Ph. D. students. Prof just stepped in to show how it's done. And he's good
Darn, looks like I'm the closest sysadmin sub while our admin is on vacation. #whatarecomputers
Replying to @superglaze
I only hope the always on feature is more resilient than on the xb360. There, you're constantly disconnected often.
Whoa, what happened to the icons in the Twitter app on Android? Well still better than all G+ UI "improvements"
Media
Replying to @fhuszar
but I agree. Definitely would require at most 3 months of winter. Not 6.
@haiqus what I'm wondering is is what "is like" is supposed to mean. It seems a lot of human level AI is going on there.
Replying to @superglaze
if I ever get one (and I know I will) I'll make sure to put it behind a switch.
@haiqus another "formula" to describe startups and projects is "X = Y + Z", for example "Spark = Hadoop + Streaming" or so.
Ok, X is like Y for Z seems to mean that you take all properties of Y and replace the one which is in the same category by Z.
Replying to @ChrisDiehl
that's comforting ;) have you Googled him. He's probably the Trevor Hastie of another community.
Replying to @noorisk
I'm sure he had his reasons, as a Scala user myself I was just interested in whether I should be worried. Still don't know... .
Can someone with a linguistic/philosophical background comment on what kind of statement "X is like Y for Z" really is?
@gappy3000 I think it's just going to be a parking lot. Although japanese garden would've been nice ;) And much quieter!
Achievment "Ignorance is bliss" unlocked: Asked a guy with a five page wikipedia page what he's doing.
2
Oh thank you autocorrect, the one time I get everything right you step in and ruin everything.
What, Twitter Germany's new Music & Entertainment Manager has only 217 followers??? #imnotmakingtherules ;)
Replying to @meltomene
@melioriot and there are other options like textbooks, playing around with stuff, talking to people, etc.
Replying to @meltomene
@melioriot oh yeah, didn't say they are not useful. Just meant that they are as good (or bad) as any form of instructional material.
Replying to @meltomene
@melioriot yeah, MOOCs with exercises and all are a different story. Although I like the one-on-one interaction you get from a real tutor.
Replying to @meltomene
@melioriot which is a bit short-sighted IMHO. They can be an excellent source of information, but you still have to do the actual work
Replying to @meltomene
@melioriot yes, I agree, of course. I just occasionally meet people who are like "now there's this course, now I can finally learn it!"
Outside looks as if someone replaced all the light bulbs with 10W energy saving lights. #winteriscoming
2 1
@giures certainly. They might even be good to get first hand knowledge and insights. But so is a good textbook.
The truth is, to learn new stuff you need to wrap your head around what you cannot understand. It takes work, no short cuts exist.
1 3
Given how ineffective I found lectures compared to self-study, I'm not really feeling the current excitement with online courses.
2
Replying to @ayirpelle
somehow I seem to have begun to assume every handheld devices is always online.
1
Just caught myself thinking the TV remote was silently flashing to show some notification. Is there a name for this? #internetofthings
RT @DataKind: Pro tip from @jakeporway: start with the smallest task you can think of before tackling the big problem. #datadive
But in a way, ML is exactly this: let a computer infer connections which are hard to engineer by hand.
2 1
Replying to @pavlobaron
how many levels? Three or more => you're in deep trouble. But you probably already knew that ;)
Replying to @pavlobaron
darn. And then you realize the counter is off and it's really 217 slides.
Replying to @pfleidi
yes, that and logging and caching and notification synchronization across devices.
RT @esammer: @michbarsinai @dcsobral @jamesiry @mikiobraun @jasonbaldridge if you really knew c++ you wouldn't want that feature.
RT @esammer: @mikiobraun @jamesiry @jasonbaldridge I already wrote a blog post about how that feature is what's wrong with our industry.
Replying to @jasonbaldridge
@jamesiry You might not have heard of that feature, I wrote it as part of my Ph.D. thesis.
2 3
RT @jasonbaldridge: @jamesiry @mikiobraun Poser. I understood it while it was still in the ivory tower. #retrohipster
RT @huitseeker: @jamesiry @mikiobraun @jasonbaldridge In other words, you got it while it was hot ?
European Internet would probably mean you need to have local servers (generating taxes) and dedicated intercontinental lines (costly).
This whole discussion about a European Internet is driven by ulterior motives besides privacy.
Replying to @scheidegger
@cartazio I think if you need a math book on probability theory first you can fill in the gaps (and despair about the errors).
Traveling from Charlottenburg to Kreuzberg can't be so much different than going from Manhattan to Brooklyn. #berlin #yeah
2
Replying to @GenKnoxx
as long as you don't hack the mail server to discard emails colleagues claim to have sent you this seems fine. ;)
1
Replying to @StephanNoller
;) Really like how they are downplaying the implementation. "simple extension" but "non convex opt problem". Hm.
Replying to @chrshmmmr
this should be part of learning. Generalization and abstractions are the same thing which give you concepts to talk about things.
Replying to @aCraigPfeifer
3.0 - 3.2 to be exact. They should've stuck to a point release = version name scheme.
Replying to @chrshmmmr
Also, we need some notion of abstraction & generalization beyond the purely statistical sense of "perform well on new data"
Replying to @chrshmmmr
hard to say. What's missing is some ways to encode "meaning" beyond bag of words & vecs. Dependency parsing looked interesting.
Just realized that Andorid versions 2.3 = Gingerbread, 4.0 = Ice Cream Sandwich, 4.1 - 4.3 (!) = Jelly Bean and 4.4 = KitKat. #whatamess
1
RT @munterluggauer: @mikiobraun use the enthusiasm of students to find out what works and what doesn't. let them understand why in order to…
Replying to @munterluggauer
I really like the second part of your reply. You are right, we shouldn't infect young students with our sarcasm ;)
Replying to @munterluggauer
what do you mean with "there"? The disappointment? Or the realization that we're don't know how to do it yet?
Replying to @munterluggauer
But over time I've learned that it's all statistics and tricks with numbers.
Replying to @munterluggauer
I believed that it could be possible to build algorithms that begin to understand data the way humans do.
RT @noelwelsh: Ladies and gentlemen, we have a winner! :-) “@mikiobraun: @noelwelsh maybe both? ;)”
Ok, hard problems in CS, updated: caching, naming things, logging, and synchronizing notifications.
Replying to @cartazio
yes, probably much cheaper than $1000. Then you need a few switches and many USB ports ;)
Replying to @jonoandre
but then you can get started with any of the text books like Elements of Statistical Learning by Hastie and Tibshirani.
Replying to @jonoandre
that's a hard one. I've become convinced that you need a working background in math (analysis, linear algebra, probability theory
Replying to @cartazio
even if you assume $100 for one of them, you can get about 30 for $3000 worth of apple hardware.
Replying to @jonoandre
it works, but if you study it long enough you see it's just magic tricks with numbers. Ah, I'm being overly sarcastic, sorry ;)
Replying to @StephanNoller
they probably believe what they are doing is a good use of data. Finding the bad guys, etc.
Replying to @StephanNoller
because you are in the UK and the other half of the people is from the US?
I wish I were that young again. Unsee the things I've seen and unlearn the stuff I've learned to get that faith into machine learning back.
1 4
We've got a batch of new students who discuss things like whether computers will be able to compose music as good as humans.
Replying to @JackTWoolford
I encourage you to go forward. Maybe you will succeed where we all failed!
Based on these revelations the carelessness with which people use mobile phones (and Skype!) on Homeland seems quite unrealistic.
So GCHQ are targeting innocent IT people in order to infiltrate infrastructure and gain access to any mobile device BASED ON WHAT EXACTLY?
Replying to @karpathy
same here, triple digits unread count. Also massively impresses any bystanders. "I don't even have time to mark emails!!"
1
Eventually, you realize no amount of wiki, chat, to-do list will help if people don't do what's necessary to solve the problem.
3 1
There seems to be phase in the life of any computer scientist where he believes the solution to any organizational problem is technological.
2 3
Replying to @andyrtd
nope. Actually I was drinking neither before I got kids, but no I'm running a solid 2-3 cups per day caffeine diet.
What I learned from giving the big data talk yesterday: apache project names and logos are just silly. I got request for more animals.
I hope that 2.6GB download for the Windows 8.1 update is worth the wait. Ah, who am I kidding *sob*
1
Physicists sure love PowerPoint. Which makes for some ugly formula typesetting. At least it's not set in comic sans, er, the Higgs font.
Currently listening to Alexander Tkatchenko on the state-of-the-art in computational atomistic simulations. Quite lost TBH.
Replying to @gehrigkunz
@gehrigds @PlanetCassandra actually we stopped using Cassandra two years ago and switched to custom made stream mining algorithms.
The general premise for any scientific talk should be that you've lost all your audience five minutes ago.
3 2
I don't even know what a European Internet means. You need to have servers on EU ground if you have European customers? How could you tell?
1
Replying to @munterluggauer
I think there is just so much wrong with what the politicians say, starting with the idea of a European Internet.
Replying to @munterluggauer
definitely. Verheugen said some very strange things like "companies can't do this on their own, so we need a big consortium"
German politicians are talking about a "European Internet" again, cite Airbus as a success model. Please let this just goes nowhere...
Replying to @munterluggauer
@arusbridger indeed. Like almost all of Twitter... I hope the UK begins to realize what's really happening in their country.
1
@thinkberg yeah and they didn't really deliver in that respect. All the great next gen games launch only next year.
Our verdict, however, was that we need to wait for some awesome games. Technology wise the next gen oomph is missing somehow. #XboxOneTour
Getting all psyched about what awesome giveaways they'll have at #XboxOneTour, e.g. Xbox one tour avatar t-shirt xbl codes.
Replying to @jakubzavrel
oh yeah, I was solely referring to the marketing work, hence the word "franchise".
Not that I question the usefulness of deep learning, but people tend to interpret so much into what these algs do by analogy it's amazing.
1 2
And security people always told us we have to fear email viruses, key grabbers, and botnets most. I feel kinda misled.
Another round of patent wars around Android? They should just fight it out with a good ol' round of rock-paper-scissors-lizzard-spock.
1
Replying to @truemped
That's really a loooot of money. TBH, I think their biggest asset is their investor network. No one else could pull this off.
6Wunderkinder about to close a $30M round. For a to-do app. (╯°□°）╯︵ ┻━┻
2 1
RT @gregor_riegler: If you are using #Linux as an OS for #Programming, please retweet this. *experiment*
Why do programmers confuse Halloween and Xmas? Because Oct 31 = Dec 25! ;) #SCNR #haventseenthisyet
17 3
Replying to @mpompery
@zeit_geist instead, it should say "oh I'm on a mobile network, maybe it doesn't work all the time, I'll just try later"
Replying to @mpompery
@zeit_geist I see. What I find irritating is when it tries to send the msg right now, but there's no internet and it throws an error.
Replying to @mpompery
@zeit_geist using chat with XMPP all the time. I do get duplicates when using EDGE or so, but that could be fixed with random IDs per msg.
What's really embarrassing is that there are still things on the Net which @thinkberg has to tell me about in person.
What's really shocking about the latest #NSA revelations is that they have also copied my hand drawn slides style!
Media
3
Replying to @pavlobaron
@jamesgolick just give it an hour. But don't forget to take an umbrella with you ;)
@moellus immerhin, hier gibt's so ein Kindersendungentitelmelodiemedley. Drei Durchläufe à 19 Minuten auf einer CD.
Maybe my biggest contribution in the talk will be a nomenclature which allows every Big Data project to describe as "X is like Y on Z".
1
Hm. My first list has 20+ projects on it. Also Cascading and Cascalog are not the same thing???
Starting to pencil my upcoming Big Data talk for our groups retreat. Probably should have called it "The History of Big Data".
1
OH: "Distributed computing with immutable data structures is like couples which have stopped talking to each other."
1
Replying to @Major_Grooves
@Ryanair oh well, you just gotta love Ryanair's awesome service...
Replying to @fhuszar
They = people trying to get into data science. I meant data analysis principles, but also infrastructure/scalability principles.
Replying to @fhuszar
They quickly get bogged down in complex frameworks and never get to really understand the underlying principles.
Replying to @fhuszar
I agree in principle, but I've seen too many young people behaving like kids in a candy store.
I know I'll sound like I'm an old guy, but the problem with Hadoop et al. is that people never learn to write stuff from scratch anymore.
Replying to @bigdata
sorry we couldn't make it. Good to see some movement in the approx algorithm space, though!
. @thinkberg @vesselhead I like this line best b/c the ASCII art: " 285 acknowledged writes lost! (╯°□°）╯︵ ┻━┻" #tableflip
1 1
Quite nothing to get your system going at the start of the week than getting two kids ready for kindergarden and school.
When I was younger I would make up stories about Bernoulli and Fibonacci on the spot. I was basically telling math fan-fiction.
2
Still kinda amazed that Angela Merkel made "das geht gar nicht" official. #NSA #merkelphone #vertrauensbruch
1
End of daylight saving times before kids: yeah, sleep in on a Sunday. Now with kids: oh no, Sunday will be one hour longer.
1 1
Just learned about LinkedIn's Intro. IMHO it's an intersting idea but the technical realization via email hijacking is just totally bonkers.
1 1
Whenever I talk to one of my academic colleagues about startups I realize how far I've come.
1 2
RT @mgdm: Here’s why LinkedIn’s email hijacking thing is nuts. Someone stopped going “noooo” long enough to write it coherently http://t.co…
I'd rather see a course on data science which shows how far you can get with the simplest stuff on actual applications.
4
RT @markusandrezak: If they listened in to Merkel, they did it for the higher purpose of terror prevention and I guess she’s got nothing to…
RT @markusandrezak: Merkel needs Fukushima to get the potential of Nukes and she to get listened in personally to get the potential of the …
Replying to @fhuszar
thanks! Look interesting. I think I have heard about Sean Owen before... .
Anyone else getting offers for SEO consulting? Are they any good? They promise quite impressive improvements. #howcanpeoplefallforthis
Replying to @PhilDarnowsky
yeah, or they get stuck for 20 minutes on update 13 out of 37.
1
Always amazes me that a Linux distribution upgrade can be performed on a live system while Microsoft security updates only work on shutdown.
1 1
Replying to @superglaze
I feel your pain. And it's never just this but countless reboots to follow... :(
@moellus habt ihr jetzt schon professionelle Schauspieler, die Schlüsselszenen aus dem Büro reenacten?
RT @johnmyleswhite: @debasishg Glad you liked the talk. The SGD needs more love in introductory ML courses.
RT @headius: "Don't define yourself by hating other things. Define yourself by building cool stuff." - @steveklabnik at #dotrbeu
We just composed this haiku to praise the internet: "legendary beast take form in present day world you grant joy and bliss".
Replying to @peteskomoroch
@joe_hellerstein ;) make sure you add Jefferies tubes so you can send interns into them.
So looks like the Republicans ended up with nearly no concessions at all. Well, if you're going in with all you got that sometimes happens.
Construction works outside our window seriously sound as if whales are about to do a landfall.
Somehow Google analytics just gets more and more complex and unusable with every iteration. Either I'm not enterprise-y enough or...
1
The best thing about tumblr is that reblogging is so easy you can generate content with your cerebellum only.
1
That My Funny Valentine live album from the Miles Davis Quintett - How can music be so intense and so calm at the same time.
Replying to @alexia
ah, good, for a second I thought you were talking about @arstechnica ;)
RT @janl: Why local routing won’t help fight surveillance. Great one by @superglaze where he dismantles a Deutsche Telekom plan http://t.co…
Replying to @superglaze
can someone whip up an infographic showing of how many components a web page consists? Politicians really seem to have no idea.
1
Googled for "batdad vine", autocorrect gave me "Bagdad vine". Now I'm worried some NSA keyword trigger went off. #hesreallyfunnybtw
Replying to @pablochacin
the weather doesn't really change down there, does it? ;) we could compete in summer but now it's more like 7/13 and rain :(
Replying to @NicholaAbdo
@Nichola_Abdo "while(not converged)" is also a good one. Fingers crossed it actually does ;)
What is it with the Ruby community anyways? Never seen a community self destruct in that way again and again.
Replying to @alansaid
Oh great, so you can use the conference's afterhours to finish the paper?!
RT @alansaid: Another sign of the lack of relation between #recsys and #IR: @ECIR2014 paper deadline is during @ACMRecSys #recsys2013 #ecir…
So today the university is taking half our offices from us. Maybe formally justified, but the process felt quite differently.
@s1m0nw on the rain radar it looks like a wider rain band is approaching from the south. Maybe it has discharged till it gets here... .
Oh great, South Bavaria already reporting record snowfall. More than ever before since beginning of records in 1800. #winteriscoming
1 1
RT @lawrennd: I've resigned as Associate Editor in Chief of IEEE TPAMI because I think it is wrong to charge $1750 for #openaccess.
Also reminded me how statistically uncommon it is to not fear math at all. I only fear badly written papers, that is all ;)
2
She also laughed at me when I told her I work at a university. "You're still at university???" - "Yes but I do research and stuff...!"
1 3
My 6yo asked me yesterday whether I'm "working math". If that helps to instill faith in her that math is doable, I'm happy.
1 1
Replying to @tyldurd
My personal favorite is Speak No Evil. With Herbie Hancock and Ron Carter replacing Tyner and Workman, but still Elvin Jones on dr.
Replying to @tyldurd
they're all golden. I think I actually own Juju. Should go back and listen to it! ;)
The True PostDoc doesn't let his student redo plots if the font size is to small. He directly edits the plots.
2
Replying to @karpathy
I just love math jokes. Always make me loose a couple of followers. ;)
Often in Big Data, it's less about the amount of data and more about the amount of time you're willing to wait.
2
Replying to @sscdotopen
of course, it's more the classical number crunching "version" of big data.
1
Replying to @sscdotopen
actually, matlab already has parallelization stuff in it like distributed matrices.
1
It's that time of the day where I have to write some Word docs. Time to fire up that ol' Windows VM. Oh it wants to update & reboot. #always
Replying to @alung
sometimes they also just relabel whatever they've been planning to do... .
Replying to @alung
at least MapReduce? Sometimes they like to keep it abstract to make the knowledge last longer.
1
Replying to @alung
actually, we have a cluster with about 20 nodes or so, but it runs gridengine, with shared filesystems. The opposite of MapReduce ;)
1
Replying to @peter_c_william
I can see why, but Hadoop is becoming very relevant fast so I feel bad for them not learning anything about it.
1
Believe it or not, most students in ML have had no exposure to Hadoop or other Big Data frameworks.
4 1
Replying to @sscdotopen
don't think so and it's going to be somewhere in Brandenburg. But I'll happily share the slides.
Agreed to give an overview talk about the Big Data software landscape at our group retreat. This is going to be fun.
Replying to @Nico
einziger ICE halt zwischen Köln und Frankfurt. Im Wechsel mit Montabauer.
Just learned about "experience rating". Have they never heard of the law of large numbers a.k.a "production law of the insurance industry?"
1
Replying to @pavlobaron
I'm not at TED level, too ;) But many seem to like to emulate that style.
Replying to @pavlobaron
and I agree. Too much focus on presentation, details and the actual quality of things gets eroded.
1
Replying to @pavlobaron
was just asked to give a talk. Requirement: 5 slides top, at most 16 words per slides, 30 mins. Seems pretty TED-y to me...
1
Replying to @pavlobaron
I wish I were that close ;) Currently it's more like "I knew he would be playing that!" ;)
Replying to @pavlobaron
what happened, have they just told you you're about to be replaced by an actor? ;)
Just realized I know every single note of the solos from Wayne Shorter's "Witch Hunt." Ah, I had so much time in my youth.
1
Replying to @brendan642
@gappy3000 yeah, when you're algorithms are O(n²) or worse, a few gigabytes should be enough to keep you busy.
1
Shouldn't have complained about public transportation earlier. Now my train is running late almost 2h. Where is Mehdorn when you need him?
So, America, seems like Italy managed to juuuust avert a bigger crisis, how are you coming along?
Replying to @isaach
cool. Wonder how well they'll be handling languages with more complex grammar than English.
Replying to @pavlobaron
dude! Sometimes it's OK to nap for like fifteen minutes. That really helps.
Replying to @blattnerma
just something that crossed my mind during one of those startup pitch events last week ;)
@hi8ussouth you shouldn't blindly favorite everything which has the word participatory in it!
Son, we live in a world that has funds, and those funds have to be guarded by men with Ph.D.s. #AFewGoodProfs
1 1
Me: "that 800 tweets in timeline limit is becoming a problem for me" Him: "Drinking from the firehose, eh?"
Replying to @noelwelsh
@fhuszar increasingly I believe that there's a honest, serious way to deal with partitions, ... and there's eventual consistency.
1
Replying to @mdreid
:D awesome! I'll show it to said student later. He's not even on Twitter... Ah, those kids... ;)
Replying to @mdreid
one could even say the horrible parsers in text adventures lead me to being interested in AI.
1
Replying to @cartazio
yeah its OK, I guess. It's not like people had gotten my jokes before ;)
Just talked to a mid twenties student who didn't know about text adventures as a gaming engine.
IMHO that recent "return of Java" story on Wired is another attempt to tell a story at any cost to create a false sense of momentum.
1
What's with all that nonsense about "the return of Java"? It was never gone server side and it didn't return client side.
1 1
Replying to @alansaid
which is really ironic (and annoying) given that they are such a data driven company...
Replying to @alansaid
why am I not surprised? Sometimes I wonder whether they have any customer facing support people at all.
Replying to @sscdotopen
@muratk3n ah, come on, you two. Let's have a coffee when you're back, @sscdotopen.
Replying to @PFCdgayo
you're right, but somehow it is not practical if everyone asks all they didn't understand, so there's a limit you don't have 1:1.
Replying to @PFCdgayo
I already mean basic interactivity like asking to understand what is unclear.
Replying to @DRMacIver
yeah, I know :( how about ten common followers from the same connected component?
Replying to @DRMacIver
but I know exactly what you mean. Mentions should only register if you have more than ten followers or other social graph guards.
How do you measure the effectiveness of a talk? How large does the audience have to be to make up for the lack of interactivity?
Besides that the usual suspects: reinforcement learning, robotics people believing in embodiment, and a few vanilla ML people.
@muratk3n Also, it seems to do very little caching besides what the OS does for disk access.
@muratk3n Postgres was pretty solid IMHO, unless you constantly add data while doing queries ;)
Discussion on representation learning challenging to follow because of a lot of reverb in the room. Oh the irony.
So today I gonna learn about the current state of the art in autonomous systems research in Germany. #subbingformyprof
1
Replying to @markusandrezak
Jo, bis 4.9% ist alles safe, danach sollte man sich wahrscheinlich schon sicher sein. ;)
@holadiho @BusinessPunkMag hehe, ist einem Freund von mir mal in einem Bewerbungsgespräch passiert. War dann nicht mehr sooo interessiert.
Replying to @glouppe
@jakevdp um, I'm actually not organizing this year. But @ahonkela should be able to help you out.
Poster at the Big Data Workshop at TU. John Hopcroft is supposed to attend but @thinkberg just told me I missed him.
Media
2
Replying to @fmueller_bln
I prefer to take an entirely behavioral point of view of politics ;)
So election: Basically, it's about voting for weights on a hand full of policy priors to solve our vast array of problems?
Also in the CLANG update: "once you have taken a bunch of people's money to do a thing, you have to actually do that thing" Well, duh!
I think the Nokia excursion paid out very well for Elop: hardware gig on his CV, $19M in cash, and he can return to M$ in a better position
Replying to @PFCdgayo
nice post! One question: in the 2nd paragraph did you mean Tps is a mostly meaningful or meaningless metric?
Replying to @mpompery
@zeit_geist 80% of the time I'm wondering whether ads on Facebook are actually real or fishing attempts.
RT @NeinQuarterly: There's a fine line between German attention to detail and German obsession with detail. An extremely precise fine line.
At least there is more diversity in ads on Twitter than on YouTube. And don't even get me started about Facebook.
Replying to @cartazio
good. But me not agreeing with you should not make you question your sanity ;)
Replying to @cartazio
yeah, you're definitely right. Getting both right is the actual challenge!
1
Replying to @cartazio
very well! I guess you took care of the fast algorithms part first, then ;)
1
Replying to @andyrtd
I see. Yeah, in that environment you'll certainly end up with something flexible because there is so much structure to plug into.
Replying to @pavlobaron
yeah, "pressure" is something completely different from a piercing pain somewhere near my brain... ;)
Replying to @pavlobaron
yeah. I had a dentist who said "there might be a bit of pressure" whenever he put a sharp, pointed object in my mouth.
Replying to @pablochacin
thanks ;) Can't quite give in yet. After all, this isn't even Winter yet. How am I supposed to survive when it's -10°C... ;)
Replying to @pablochacin
during summer Berlin could hold up, but I guess that game is lost. Barely 6°C in the morning now, Brrr.
Ok, through which administrative or political organization's ranks do I have to rise to move school's start from 8 to 9am? #yawn
1
After staying home looking after my sick son went to work for just one meeting which filled up the whole slot. Maximum efficiency.
Replying to @InkmiHq
@codemonkeyism I thought the Q10 at least had the classical formfactor with physical keyboard. Still no good?
Replying to @DRMacIver
as I see it, unicode is ok, the encoding stuff is what bites most people.
Clearing all notifications after you watched the item seems to be a really hard problem for all social networks.
@muratk3n sure, definitely in speed. ;) Also, resource usage is very predictable which is always good.
Replying to @alung
ah well... . Also shows how the Big Data hype slowly disconnects the buzz from the what's actually happening... .
Replying to @alung
what bugs me is that they're already becoming that stream ML white card in discussions like "yeah, but have looked at mlbase?"
Replying to @alung
there was just some buzz recently about mlbase for Spark, but what I saw didn't really impress me yet.
Replying to @alung
yeah, definitely. I think all the "classical" ML toolboxes do this right.
IMHO the real challenges with ML libs are not implementing a bunch of learning algs but tying them together with data handling and preproc.
5 4
In particular what's missing: Strong preprocessing stuff, meta stuff like cross-validation.
@muratk3n in particular this: "[loyal Apple customers] remained quite happy with the previous model, discontinued in 2011."
Really funny how Amazon's Auto-Rip makes CDs more attractive sometimes because they're cheaper than the MP3 version.
Replying to @huitseeker
.@huitseeker I really meant the web, that crazy stack of HTTP + HTML + CSS + JavaScript + browser plugins.
. @thinkberg "He said that founders should listen politely and just do what they want to do anyway."
Replying to @sscdotopen
@zeit_geist ah, I see. Not sure whether people are aware that you don't see Mahout as Hadoop-centric.
Replying to @sscdotopen
@zeit_geist it's not? How is it done? Do you use some intermediate abstraction like summingbird?
Replying to @mpompery
@zeit_geist ah, new blog post coming up: "3 different reasons why you want to scale up, size, latency and X"
Replying to @mpompery
@zeit_geist computation up as much as possible, possibly even based on matrix algebra.
Replying to @mpompery
@zeit_geist yeah certainly. But often you try to learn from just a few 100s of GB and it already takes a week. You'd need to scale \
Replying to @mpompery
@zeit_geist that's why I think matrix algebra is the wrong abstraction. You don't really "see" whether something decouples nicely or not.
Replying to @mpompery
@zeit_geist actually I think you'd also need new ways to think about learning algs which stress locality. Right now it's too accidental.
Replying to @mpompery
@zeit_geist you mean it's all there it only takes a clever mind to integrate it all? You're probably right ;)
Replying to @mpompery
@zeit_geist that depends a lot. Often you introduce approximations to improve data locality.
Replying to @mpompery
@zeit_geist put differently storage isn't really the problem but computational time. (think O(n^3) algs)
Replying to @mpompery
@zeit_geist streaming the data past the model. Other algs like community detection in graphs need to take in all of the data.
Replying to @mpompery
@zeit_geist hard to say without discussing specific algs. What matters is the complexity of the learned model. For SGD, it's more about \
Replying to @mpompery
@zeit_geist I think part of the problem is that ml itself is pretty scarce in computational primitives (besides matrix algebra).
Replying to @mpompery
@zeit_geist storage is probably too close to classical infrastructure problem domains.
Replying to @mpompery
@zeit_geist TBH, classical ml frameworks also leave me flat these days. Haven't yet figured out why. They are often very generic.
Replying to @mdreid
yeah, don't know what it is. I feel I get a much better picture of a guy from his Twitter profile & timeline than on G+.
RT @huitseeker: @mikiobraun Oracle is like a rock star on a retirement tour. Just one more ...
Somehow, people on Twitter feel much more real than on G+, in spite of real name policy and more than 140 characters.
@pbrane except for the fact that distribution is not the only way to deal with huge amounts of data, sometimes better algorithms help, too.
.@thinkberg trying to explain me that Apple have no other choice given differences in VAT and import tax. #yeahright ;)
Replying to @captaink99
Probably should add that I owned an HTC Desire and a Sony Xperia mini pro before that, so competition ain't that fierce. ;)
Replying to @noelwelsh
I only envy iOS for their awesome music apps. And the 3D view in maps. But that's about it ;)
That being said, my Samsung Galaxy S3 LTE is by far the most awesome smartphone I ever owned. And it cost a bit over 400€ without contract.
RT @headius: Google Summer of Code is about students learning, not doing free work. We pass students that accomplish nothing for JRuby but …
Actually I was hoping some people would come forward after yesterdays and say "Haven't you tried project XYZ? That's exactly what you mean".
10°C in the morning? Berlin what the heck? Why do we never have extended periods of comfy mid 10s in fall?
Replying to @mpompery
@zeit_geist maybe I'm not cutting the camps along the right dimensions, yet... .
Replying to @mpompery
@zeit_geist yeah ;) I also wrote "SQL" in the context of Big Data and later thought "whoa" ;)
For the rest of us, the latest NSA revelations are without immediate consequences, or are they? Which makes it all the worse.
There are a few professions where always assuming the worst is mandatory. One example are lawyers, another one is computer security.
3 1
Replying to @dwf
but we have elections in a few weeks. Let's see who's in charge after that. I don't get my hopes up, though.
Replying to @dwf
yeah, it's really about THE INTERNET! I don't know, require Google to process German communication locally, perhaps?
I'm not even sure what mental model some politicians seem to have of the Internet. Telephone? Mail? Postal carriages?
3 1
Ah, I forgot, we're using the US Internet right now (Google and stuff, you know). Would probably take some time to build a German Internet.
RT @andyrtd: @mikiobraun Would like to read your blog post. Just issued a SQL statement for training a Decision Tree model... :)
Replying to @mpompery
@zeit_geist I should call it "join vs. matrix multiplication". Or are they even the same thing? ;)
The chasm between database guys and machine learning is deeper than you'd think. Different sets of abstractions and methods. Should blog...
2 3
People need to be more aware between infrastructure type Big Data and tools which can readily produce real value.
RT @janl: If you think open source crypto is more verifiably secure for the average nerd, you don’t understand who contributes to open sour…
Replying to @mpompery
@zeit_geist then again, most people probably wouldn't notice the difference... .
Replying to @mpompery
@zeit_geist yeah, it's certainly safer that way. ;) Although I think the Dropbox+Apps way would probably work in the market.
Replying to @mpompery
@zeit_geist yeah, and fiber channel to the home! ;) for me, could also run on a VM as long as everything is encrypted.
Replying to @mpompery
@zeit_geist @janl App ecosystem like on smartphones. After all, a smartphone is already a linux box + apps people understand.
That whole (consumer) cloud idea seems increasingly difficult. Access to your data at all times is nice but the loss of control... #NSA
1
Replying to @squarecog
@noelwelsh had a look with summingbird. So how would you access the results in real-time of an ongoing computation?
In Germany, memories of misuse of surveillance are still fresh whereas in the US and UK it helped to win WW2 and fight terrorism. #YMMV
2
But isn't this attack on cryptography exactly as in the rant towards the end of Good Will Hunting ?
So far my attitude towards basic auth and weak crypto was that nobody was interested and it would be too much work. Looks like I was wrong.
Replying to @Nico
meine Tochter sagt gerne mal "lach lach". Immer noch besser als "LOL." ;)
@muratk3n it _is_ a pretty lazy name. Like "pq-formula." But I guess it's your right of you're the first ;)
@muratk3n they were probably thinking QR stood for something like "quasi reduction" or so ;)
@thinkberg Sometimes I think that's the main purpose of Silicon Valley: it's just one big dating pool for co-founders.
Replying to @noelwelsh
also, my biggest gripe with Storm et al. is that there is no "query layer", it's really just computation.
1
Replying to @noelwelsh
yeah, it's really "just" a piece of infrastructure. That alone doesn't help you to do something useful with the data.
Google's infrastructure papers are always a good read. I like how they design systems so that reasoning about it is as easy as possible.
2
Read the MillWheel paper (Google's stream proc framework). What can I say, yet another framework. Guarantees seem to be nice, though.
3
RT @GaelVaroquaux: Writing yet another grant to try to fund @scikit_learn. Companies using it, help us! We need testimonials http://t.co/Pg…
Replying to @ian_soboroff
@ChrisDiehl I also like quite didn't get that minimize over hidden representation part TBH
Replying to @ChrisDiehl
he is the master of project meeting, the only one never to do email and always stay on top ;)
Replying to @ChrisDiehl
So, why not just do brute force and fit some function based on a shallow model with 100k parameters and proper regulariztaion.
Replying to @ChrisDiehl
invariants in your data, find representations which encode "meaning" in independent features. But the search space is too large.
Replying to @ChrisDiehl
considered is just too narrow to be powerful enough. My personal view is that finding good features amounts to learning \
Replying to @ChrisDiehl
IMHO, MKL are just some way of doing cross-validation more or less on-the-fly, and the set of possible feature representations \
Replying to @ChrisDiehl
hm. TBH, getting good features requires to understand the data, something beyond manifold learning, feature extraction, etc. \
Replying to @ChrisDiehl
man, I've sit through so many hours of project meetings with this guy.
Replying to @ian_soboroff
@ChrisDiehl Note that since SVMs are single value output, you'll need one SVM for each hidden layer node. Trained with SGD.
Replying to @ian_soboroff
@ChrisDiehl They patched together two layers of SVMs and did some min-maxing over the representation in the hidden layer.
Replying to @superglaze
hehe. Well at least it's not like he brought Nokia back to the top 3 smartphone brands.
After Elop's stellar performance at Nokia, he is the natural choice for heading the new devices division at Microsoft.
1
Listening to a talk on adaptive filtering I'm thinking "all they're doing is prediction!" Same could be said about machine "learning" ;)
1
Replying to @mdreid
@ChrisDiehl that there is an amount of feature abstraction going on comparable to what the human brain allegedly does.
Replying to @mdreid
@ChrisDiehl I have co-authored a few papers on internal representations of deep NNs, but I'd say it's still a stretch to claim \
Replying to @izendejas
switched to disconnect because I didn't like adblock's blackmailing policies... . Well, you can't have both it seems.
1
Replying to @ChrisDiehl
@mdreid minimize the dual objective over the hidden-layer features. Looks suspiciously like ML marketing. Maybe put some Bayes \
Replying to @ChrisDiehl
@mdreid Hm. Not sure why this should do anything different from "normal" deep NNs. I also don't quite understand why they \
You know you're building a framework when you introduce wrapper classes for all basic types.
Ads on @YouTube weren't that immensely annoying if they'd show me more than those same six ones over and over again... .
Replying to @cartazio
yeah, man. Was fun chatting with you! I'll be off to work (although I wouldn't mind sleeping some more, too)
Replying to @cartazio
it can work, but only if the editor is sufficiently engaged. Too often he's not.
Replying to @cartazio
one of the editors had a paper where he proposed design patterns for FPGAs, that got me interested.
Replying to @cartazio
wait, how old are you? ;) just kidding. That was the C64's younger sibling. Probably only successful in Europe ;)
Replying to @cartazio
you know, I once soldered a RAM chip piggy back in a C=16 to extend the mem to 64k. Sometimes I miss those days.
Replying to @cartazio
Hm. Yeah it's about time I get my hands on an ARM processor ;) I should also read that FPGA book some more.
Replying to @cartazio
well it's too late for that now ;) just like I dropped non-SSE3 support in jblas. I just ran out of systems to test it on.
Replying to @cartazio
nobody cared (me included). It was like Flash/Silverlight based on Java. Like applets with cooler visuals. Huge waste of time.
Replying to @cartazio
yeah, the Sun erm Oracle bug database has open bugs which could go to school...
Replying to @cartazio
the main problem is that the JNI was designed as a vendor independent API. And they didn't left enough room for perf hacks.
Replying to @cartazio
that's good. I always get the feeling that the Java NI has extra penalties for moving out of Java land.
Replying to @cartazio
@onecreativenerd great! I don't think I follow him already. That can be fixed.
Replying to @cartazio
good. We have some customers who are about to do some pilot projects with us. That is if they finally get around to deciding ;)
Replying to @cartazio
yeah, I know. It's like all Haskell hackers also need to be good at C ;)
Replying to @cartazio
I'm often surprised that CPUs are still good at floating point at all. Sometimes it feels that all it takes is being good at JS;)
Replying to @cartazio
just saw the extra scene. Does it imply everything went back to normal? #moreshivers
Replying to @cartazio
I also haven't played 1 and 2 yet. But I think it's fine to first play the latest game to get hooked on the franchise.
Replying to @cartazio
oh no. I only saw her name Anne DeWitt somewhere but only went as far as thinking she's Booker's wife from an alternate reality.
Replying to @cartazio
a friend warned me that the bioshocks are always full of plot twists, but this was way beyond my expectations.
Replying to @cartazio
awesome, I just finished it on Saturday. What a mind blowing ending! ;)
RT @BFriedmanDC: Why has no one invented a social media platform that costs $1.99/month and has no ads and doesn't track or sell its users'…
Replying to @purbon
I must admit that after the last winter I thought "no amount of summer can make up for this. " Luckily I was wrong ;)
What I really would like to see: Merkel and Steinbrück going at it at a whiteboard, see who has the better problem solving chops. #tvduell
With German punctuality, autumn arrives in first week of September. But what an awesome summer. #kannmannichmeckern
1 1
Replying to @alung
to be honest, no. There were some talks but nothing came off it. Spiegel.de has some minimal stats on tweet counts.
Best advice on prioritization I ever got is that in the end it's up to you because only you have the full picture. /cc @cartazio
Pretty impressed that Twitter rolled out the new conversation view simultaneously on web and Android.
1
Replying to @fhuszar
yeah. it's certainly related. Although we're targetting serienradar more at media companies and the like.
Replying to @mpompery
@zeit_geist you're serious? Looks like they were doing a project with Adello and they liked them very much. Very much ;)
Replying to @fhuszar
hehe. The last statement by the CTO is also a genre-defining-buzzword-ridden-awesomeness.
Replying to @fhuszar
I call it the Bieber effect. That's also why I have a little Justin Bieber sticker on my monitor.
Awesome FAQ on startups by @jaltucher. My favorite one: "21) Should I ever focus on SEO? No."
2 1
RT @endtwist: We’re squarely in the golden age of the web right now. New enough to still be a marvel & old enough to have widespread use. A…
Replying to @noelwelsh
based on ideas from statistical physics, but we never got somewhere with that approach, unfortunately.
Replying to @noelwelsh
interesting. Worst-case is so deeply entrenched in all of math... . My Ph.D. advisor had ideas about average case analysis \
Replying to @amuellerml
@t3kcit @jscastanoc @jure @h0pbeat no idea, I'm self-hosted. I use jsmath, similar to mathjax. If you can get that, it's ok.
Just remembered chatting with some guy on a BBS in '93 or so and he says "man, you type fast" all the time.
Let's be honest, the #NSA is currently running the biggest multiple testing experiment in the history of man.
And if it doesn't work, they should stop runningel those unreliable algorithms against actual human lifes. #PRISM
Actually I think we need even better Big Data algorithms to reduce false positives to an acceptable level. And with acceptable I mean zero.
1
Mann, die Potse ist heute aber auch wieder ein einziger Stau. #berlin #leidenaufhohemniveau
It enables you to deal with "web scale" amounts of data, but in the end, the responsibility is with the people using it.
1
RT @BigDataBorat: MongoDB change name to MongoDB. Will take a few day before replica sets reflect change.
Also I'm not saying review per se is bad. Paper certainly improved based on reviewer feedback. Still... . It's a looong process sometimes.
Come to think of it, don't they say citations are even more important than getting published? We should probably leave it at that. ;)
Replying to @fmueller_bln
ah, cool. It's a small world! I'll say Hi if I see him next time. He finished his Ph.D., works at Zalando now.
Replying to @ogrisel
yep. The directory where the files reside ends in 2011. Constant reminder... .
That's pretty awesome. Our Fast Cross-Validation paper has been in review for so long, we already got citations for the arxiv version.
1 3
What? I passed the 6k tweet mark without noticing it? Well, could blame it on eventual consistency ;)
I really must make people who try to explain something to me use the whiteboard more often. #visualaid
1
Replying to @KellyJayDavis
If I ever get to playing around with one of those things, you'll get the first one I fried ;)
Replying to @KellyJayDavis
Close ;) Reconfigurable Computing edited by Scott Hauck and Andre Dehon.
@muratk3n Just remembered that it was very efficient (a few cycles per pixel) and had very nice properties for all kinds of applications.
@muratk3n Unfortunately, I read the paper only up to where they described it's properties. Haven't seen the algorithm ;)
Last night I dreamt about Pozella filters, edge detection filters with many nice properties. Googled it when I woke up. Doesn't exist.
Replying to @xamat
should I be worried I hadn't heard the term Simpson paradox before? ;)
1
Replying to @msmeissn
so inhaltslose Werbe äh Wahlplakate haben wir natürlich auch schon. Mir ist aber, als wäre da immer noch mehr los gewesen...
RT @rbranson: While you kids read and write to DynamoDB, somewhere a poor soul is trying to multi-path a SAN that backs a virtualized SQL S…
Sounds crazy, but the 800 most recent tweet limit means I'm missing a few hours of tweets every night.
That 1.6% of all internet traffic sounds interesting. Where can I get it? Does gnip have it?
1
RT @doctorow: "Some 700,000 Americans every year declare bankruptcy because of medical bills. The number in Japan?..." http://t.co/adSDmwPz…
Replying to @the_real_jambi
getting old, as I said ;) No, you're right I should definitely ride the bike more often!
Completely forgot just how awfully far you have to walk at Alexanderplatz to transfer from subway to S-Bahn. #gettingold
Replying to @cartazio
ah I see. C has a very idealized view for today's architecture, access to modern features have to go through __blargh() or worse.
Replying to @cartazio
yeah, somehow C is still occupying a spot in the landscape of programming languages more or less alone.
Replying to @cartazio
oh yeah. Luckily it's mostly Intel architectures nowadays. Probably except for ARM, but haven't coded for those yet.
Reading through reviews: "I suspect you don't know either what you mean exactly". I just love the ML community. Very professional.
3
Cloud critics always said "I'd rather have my data where I can see it", but they thought their argument is hypothetical.
I don't get how program committee members ask reviewers to align their scores. People disagree, someone has to make a decisison, right?
1 1
Except for me, of course ;) Math was my minor subject. I didn't really "get" grad level math until late into my Ph.D.
The math dept in Bonn took care that there are new courses in each field of specialization each fall. And the students were insanely good.
Talking with a student just reminded me of how awesome the teaching offerings of the math department at @UniBonn were.
1
Replying to @cartazio
have you heard about usual arithmetic conversions in C? Coded for years before I first read about those.
As if the faux shopping mall experience of the transit area at Heathrow wasn't bad enough already.
Replying to @mpompery
@zeit_geist If I knew I wouldn't be thinking about better ways to do TF/IDF.
Replying to @mpompery
@zeit_geist not unless you get to give the machine the slightest idea what you are talking about.
Replying to @mpompery
@zeit_geist Yeah, you're right. Guess I got afraid at the end people would think I meant what I said. But then again... .
Algorithms don't understand sarcasm, context, or have humor, and yet we let them dictate what's considered suspicous behavior. #PRISM
7 1
If you ask me, the Rise of the Machine is already upon us. The NSA let them have access to our communication and executive powers.
1
Replying to @mpompery
@zeit_geist nicht ohne (ohwohl, soooo schlimm war das DAMALS auch nicht ;) ), aber anders! Nochmal bei TCP/IP + DNS anfangen. ;)
Replying to @mpompery
@zeit_geist hehe. Obwohl Computer sind ja ok. Nur das mit dem Internet sollten wir noch mal überdenken ;)
Replying to @mpompery
@zeit_geist das meiste guckt man ja wirklich nur ein Mal, aber manches behält man ja doch gerne.
Replying to @mpompery
@zeit_geist verstehe. Also doch kaufen. Alles Unsinn mit der Cloud ;)
Neuester Witz: lovefilm löscht alten Content. Was'n los, kein Speicherplatz mehr? Vielleicht gibt's bei der Telekom noch Cloudspace... ;)
Kann immer noch nicht glauben, dass lovefilm echt zu Amazon gehört, so schlampig wie der umgesetzt ist.
Replying to @octonion
@xamat hehe. Actually I think non-real A.I. will be even more likely to rise against its creators, so yes.
IMHO ML is mostly about how to encode relationships such that you can recover them with statistics, hopefully generalizing to new data.
4 3
Hehe, seems like Twitter's WTF algorithm cannot bring me followers at the speed I'm driving them away ;)
1
It's just amazing (and frightening) how much social context I'm reading into other people's design choices and tool preferences.
I'm going to start telling everyone Scala's proper pronounciation is so that it rhymes with the Klingon word Qapla'.
1
Replying to @fhuszar
yes. In this case it seems they couldn't resist the temptation to spice up at least the abstract themselves...
Replying to @fhuszar
reread the abstract. In particular the last claim on the effects of Facebook is so out of the line.
So far, @Medium is really great as a better-than-Google-drive scratchpad for ideas. But nothing worth publishing yet, unfortunately. ;)
1 1
Replying to @cartazio
@gappy3000 so when I do two ops like a map and a filter, does Haskel apply both ops to small chunks which fit into the cache?
Replying to @cartazio
@gappy3000 yep. I always wanted to put more work into this so that you do it only once and then run a bunch of ops in native space
Replying to @cartazio
@gappy3000 I'm thinking "this is not good for the cache". Same thing for chaining expressions in jblas.
Replying to @cartazio
@gappy3000 nothing in particular. It's just that every time I'm happily mapping, folding, and zipping large collections...
We don't need summarization, we need a service which tells whether a text contains any new information.
4 2
Anyone knows a functional collection library which re-slices computations to improve cache locality?
Replying to @mpompery
@zeit_geist there are worse things which can keep you awake at night. I guess ;)
Replying to @mpompery
@zeit_geist always in Java land ;) Yes, I know, we talked about it at bbuzz, righ?
Replying to @mpompery
@zeit_geist yeah until you use three different libraries all using different frameworks and you try to do some configuration ;)
Logging is probably one of those topics whose complexity is most difficult to explain to non-tech people.
1
@muratk3n I agree, the view, in particular to the west is awesome. But the *people*, the people... . ;) #nerdingout
Replying to @lawrennd
ok, just remember to wash your hands afterwards before you eat something, mkay? ;)
1
RT @ilparone: "Due to algorithms, demonetisation & piracy, by the 2030s, western culture had effectively ground to a halt." http://t.co/FPB…
Replying to @pavlobaron
hm. That's true, too. So you don't need summarization per se but an estimate whether there's something new in there... .
Replying to @mpompery
@zeit_geist yeah, we'll hire some frontend/design rock star with the money we raise. ;)
Replying to @mpompery
@zeit_geist no. I think we need something like snapchat for audio in a format like Twitter. ;)
Replying to @pavlobaron
I know! In particular Twitter. Only 140 characters, but linked to articles with thousands of words in them!!
@muratk3n Yes, you'd write it as separate words, but when you say it feels more like a single word. ;)
Dasgehtgarnicht = German for "that's completely unacceptable, probably violates a large number of regulations, and you need to stop now."
2 2
Replying to @fmueller_bln
hehe ;) Unfortuantely, I also don't know the answer how to make machines understand text. I'm focus on complaining for now.
Replying to @fmueller_bln
yes, yes, no offense ;) Do you have a link to one of those services (which hasn't been acquired and shut down yet?) ;)
Replying to @fmueller_bln
I'd rather read the first paragraph than trying to parse a machine generated summary.
When has the Internet turned into so much reading work? I thought we left books behind exactly for that reason. ;)
I also dreamt I had set an alarm clock just in case I don't wake up from the real time. But they weren't synchronized so I woke up early.
Storywise Twitter's WTF was a bit disappointing. It started so fresh with in-memory-on-single-server, but in the end they went all Hadoop ;)
Last night I dreamt of a new site featuring a cloud based web browser. Everyone was super-excited.
1
So school's got me again. I always hated getting up so early. Luckily I'm allowed to leave after I've dropped of my kid ;)
RT @bradfordcross: nerd out proudly. passion for your interests drives self actualization. nerding out is cool. apathy is a bore.
Replying to @quantisan
no ;) just one of those 100yr old buildings in Berlin. The walls seem to retain the heat for days even after it cools of
Enjoying the somewhat cool 16°C outside after a night of suffering in a bedroom at 27°C due to overheated stone walls.#damnyouthermodynamics
Replying to @fmueller_bln
yeah, and that was awesome ;) But what about the 45 mins of "character development" in between?
Replying to @alung
just read about the first third ;) (in line with just loosely supporting my hypotheses about the world) ;)
Replying to @pavlobaron
yeah ;) It's also a fun read. Somehow I like those industry papers.
Some days my timeline looks like everybody's just posting whatever article loosely supports their hypothesis about the world.
Pacific Rim was ok, but I don't get all the 9 and 10/10 ratings on imdb. IMHO that movies had plot holes the size of a category 5 kaiju ;)
Replying to @superglaze
yeah that was pretty crazy. At one point it sounded more like artillery than thunder.
Considering how hard it was as an undergrad to think about n-dim spaces for n > 3 it's funny that I now expect everything to be n-dim.
1
Later on it turned out that the x-axis was actually up to 3*N dimensional where N is the number of particles considered in the simulation.
"And how should we imagine the x axis?" - "Oh, the x axis is actually the x axis" - "Oh." #abstractmath
1
@muratk3n yep, that sometimes happens with those old wodden framea. Luckily it wasn't much. But that gets a Cat4 on my scale. ;)
So Nielsen (who is working with Twitter to create a Twitter based TV rating) says Twitter influences TV views?
2
First double drop-off at Kindergarten and school. This is the future. Only that it needs to take place one hour earlier from next week on.
@muratk3n oh btw, pushed new Python examples and bugfixes to SD. You need to redownload for it to work.
Replying to @SidjeRoberts
@DrNormanBurke yes, in particular diaspora. @MyCozyCloud is also interesting.
1
@gappy3000 my brain already put you and @alung into a similar cell because your profile images have similar color distributions ;)
Replying to @SidjeRoberts
@DrNormanBurke I think cloud is good, but not the way it's done right now. You need to own a piece of the cloud.
Replying to @SidjeRoberts
@DrNormanBurke And then Google decides to shutdown the service, or the startup gets acquired (e.g. Astrid ToDo app) and shuts down.
Replying to @SidjeRoberts
@DrNormanBurke I wasn't talking about redundant storage, but about putting our data into cloud services we have little control over.
@gappy3000 Dang, gotta find me another user who uses a platypus as profile image. Got kinda attached to that little bugger ;)
@zenogantner @mdreid oh, I'd be surprised if any musician used G+ (and not, say Facebook) for communication purposes.
Replying to @mdreid
hehe, I was reading Joan Baez at first and wondered "what, she is using G+?" ;)
With all that NSA stuff and one-sided service shutdowns (Reader, Catch), I begin to wonder if the push to the cloud is in all our interest.
2
RT @TechCrunchOnion: BREAKING: Google has acquired news reader startup Feedly for an undisclosed sum.
Replying to @Nico
ja, der Tag hat es in sich. Heute morgen sah's hier auch noch ganz entspannt aus... .
That whole surveillance system the NSA built is a hacker's dream come true without any fears of legal consequences.
3
Replying to @gridinoc
oh yes. Right now I'm using some by Staedler (German company?). They're ok, too.
1
Replying to @gridinoc
I personally don't like the often unclear lines they produce and the pressure you need. I ended up with fineliners, which are ok.
Replying to @gridinoc
yeah, ballpoint pens don't feel right for me for almost anything... maybe except for signatures.
If you think we'll ever get the NSA to confess the whole extent of their surveillance capabilities, you're wrong.
RT @ogrisel: I really like impl. tricks and negative results in industry papers. It contrasts a lot w/ the positive-only bias of the avg ac…
Looking into FPGAs. So it seems I've come full circle. Started with a soldering iron when I was a teen, now coming back to hardware.
RT @sscdotopen: btw, I'm in the bay area until end of september, if anyone wants to meet and talk about #mahout, contact me.
@muratk3n you, Python guy, what is the official take on Python3. Anyone still using Python2 or has everybody switched to Python3 already?
RT @shellen: When I'm on my deathbed, most of my regrets are going to be about time spent updating the firmware in things.
Replying to @tnvikram
oh yeah. Forecast for Friday is above 30°C. I know I complained about the cold winter, but this... ;)
Replying to @debasishg
yeah, in particular for things like size() which is O(n) for some where you'd expect it to be O(1).
@muratk3n ah, dang. The Twitter example also stopped working after their API level changes which leaves us without a running Python example.
@muratk3n no, sorry ;) But should translate more or less 1:1. Client structure is identical.
@muratk3n you can download a jar file and run it locally. SD page is still the beta, "real" product is coming up some time soon ;)
@muratk3n right now you would have to shard by hand. We haven't needed parallelization yet ;) And yes, you can also install locally. ;)
@muratk3n yeah, it looks at all the data, but keeps a bound on the amount of information it accumulates.
Maybe we should've called streamdrill "in-stream analysis" just to conform to the BigData nomeclature.
1
I hope there are linguists out there who study the relation between hype cycles and neologistic patterns. #doesthatwordevenexist
1
My PhD advisor used to say "finding the right model is the hard part, scaling it from days to minutes is a weekend job." Take that Big Data!
4 5
So Google apparently released another product. And all I can think is "how long till they ditch it, too?" ;)
1
RT @sscdotopen: #mahout 0.8 is released! With super fast streaming kMeans from @dan_filimon and various new recommenders from @zenogantner
RT @isaach: super post on the vacuousness of "vision" and the strength of "thesis" for a startup: "Stop Backing Visionaries" https://t.co/g…
Replying to @leonpalafox
I think there are many who still prefer academia. I also am very fond of research, but find many of the processes difficult.
1
Replying to @leonpalafox
yeah, everyone deals with it differently. Although I think many of these issues are real.
Frag mich gerade... hat jemand aktuelle Zahlen von StudiVZ? Haben die von #Prism und so profitieren können? #deutschesgoogle ;)
Gaaah they changed the Google Maps app. Honestly, looks more like Apple maps. Without the cool 3D sat view... #worldupsidedown
1
RT @gavinpurcell: Can you imagine how much fun it'd be to be in the next @syfy pitch meeting? Literally anything is possible. Earthquake of…
RT @xor: Android users: you're probably giving all your wifi passwords to Google. Who might then give them to the NSA. https://t.co/R99kpNF…
RT @dominik: I really need one of those self-driving cars. I can’t deal with all of this traffic stuff when I’m having coffee and checking …
Replying to @sscdotopen
they've also started to mail in diffs, it seems ;) Actually, it's more like :(
Replying to @sscdotopen
in effect, they're doing stuff like that by hand right now all the time it seems ;)
Crazy, Xbox 360 + 250GB HD now as low as 135€. Time to get a few and build an Xbox cluster.
2
Replying to @syhw
no offense taken ;) I just wanted to lament the fact that others made these decisions ;)
Replying to @syhw
personally, I prefer latex with git even for that kind of collaboration. Or even plain text files. But I didn't make the decisions ;)
Replying to @syhw
did I mention that the proposal I'm involved with right now is written in Microsoft Word?
Replying to @random_walker
I also love that once the bill is gone, the head is just discarded.
RT @random_walker: But the "disembodied heads on a conveyor belt delivering dollar bills to the hungry machine of capitalism" visual is har…
Replying to @pablochacin
lucky you. We were having such a great time then suddenly clouds appeared yesterday and now it's 20°C tops.
Replying to @HexstreamSoft
yeah that, too. And it's just so hard to resist this urge to generalize.
Vine is like a case study in eventual consistency. Different posts every time it loads. If it loads...
Replying to @greatparmesan
@ganeshp1990 no, Adblock Plus is allegedly blackmailing companies to whitelist their ads. I don't want that or the whitelisted ads.
Replying to @dirkhain
that whole whitelist-your-ad-for-money biz model of Adblock Plus turns me off.
@bastianventhur yeah, you also cannot really edit or whitelist some sites... Not so convinced yet... .
Replying to @msmeissn
yes, totally. Those bad adblockers! I'll look at all the ads now, that will teach them! ;)
So, should ad blocker sell whitelisting? No. Would everything be fine without ad blockers? No. There's a reason people are installing them.
I'm not surprised everybody's bashing Adblock Plus for their questionable biz model. But they're also desperately looking for a scapegoat.
RT @vivekhaldar: "We don't do email, email does us" - Gloria Mark et al have done the unthinkable: empirically observed... http://t.co/k9Yj…
Ten professors writing on a single Word file proposal with exclusive locking via email, scheduling by hour via doodle. #nowiveseeneverything
2 9
Replying to @Scott_Frye
@leonpalafox @muratk3n yeah, at that speed it gets hard to find sites willing to use all the bandwidth.
Replying to @leonpalafox
@muratk3n Interesting. So it's more due to techno-historical reasons than differences in culture?
Replying to @leonpalafox
@muratk3n we're building something for Japan right now, and they stress mobile very much. That got me thinking...
Replying to @leonpalafox
@muratk3n I'm just guessing. Do you think that's because of the academic env?
@muratk3n I think in Japanese society the community comes before the individual. Surfing on work equipment would harm the community.
@muratk3n It's probably much more socially unacceptable to do work-unrelated stuff on your work computer in Japan than elsewhere ;)
The standard web surfing scenario seems to be: on a Mac sitting in a coffee shop (US), on your phone while at work or travelling (Japan).
Apparently, you need to design web pages for mobile first and foremost in Japan because that's where people browse the web.
Replying to @roidrage
wait wait, tickets cost about 500€, prize is about 40000€. Now that sounds like a viable business model! >;)
More nouns which need to be verbed: brain ("I can't brain this" like stomach), people ("I don't know how to people").
People in the US were always more concerned about the government spying on them than privacy issues with companies. Now I understand why.
2
Twitter4J is nice and all, but keeps you at arm's length from the actual data which makes storage and reanalysis harder than it should be.
Replying to @ds_ldn
Another candidate for replacing "on the cloud" with "on the moon" for pure awesomeness ;)
Once you turn of mobile Internet and remove all apps, smartphone batteries actually last for almost a week. #gsmrules
Replying to @pavlobaron
@leonpalafox ... they would've seen the information is just not in the data, or something else is wrong.
Replying to @pavlobaron
@leonpalafox too often I've found ppl just run algs and look at error rates. If they'd looked at the data once...
Replying to @pavlobaron
@leonpalafox yeah, it's certainly flawed, but it is really good at seeing patterns.
Replying to @leonpalafox
;) an SVM which automagically choosea the right kernel for all problems!
Do whatever you can to have a look at your data. The real pattern recognition algorithm is in your brain!
4 1
Replying to @superglaze
next they'll reiterate that we need a German Google clone. Publicly funded, of course... #sigh
@bastianventhur They can always mine the URL shortener events for data, so it's probably still valuable to them.
If Google is going after all RSS based services, I wonder why feedburner is still alive... .
Replying to @random_walker
yeah, just read somewhere that that the Google Reader backend was powering other services as well (e.g. in iGoogle).
Replying to @DmitryKan
yeah sure. But usually they keep the data around much longer to download.
Replying to @cartazio
I already got my data. But two weeks seema very short. I can still get my Buzz data from 3 years ago.
Replying to @leonpalafox
and then you add the method and everything goes back to normal. More severe in languages with type inference.
RT @ikai: The funny thing about the vim vs. emacs debate is that now we just all use Sublime Text.
That moment when you type in a method you haven't written yet and the whole editor turns red like "gaah it all doesn't make sense anymore!"
Replying to @mdreid
but even with all those cool iPad music apps, learning a real instrument is still awesome ;)
Replying to @mdreid
I started with classical piano (my mothet is a piano the teacher) but switched to jazz when I was 17 b/c I didn't want to practice.
Replying to @mdreid
and getting around the fingerboard is still slow ;) So is jazz the musical style you play most on tge guitar?
Replying to @mdreid
very good. ;) Turns out the hardest thing is that so much of my piano knowledge has turned entirely into muscle memory.
All too often, the main reason for an app is to get access to the user's contact list and other personal data.
Replying to @mdreid
btw, can you tell me whether one drops the root in jazz chords on the guitar as well? You know, like playing an Fmj7 over Dm7?
Replying to @mdreid
I like the guitar. Has a stricter set of constraints, but still not too many. Piano seems to be underconstraint in comparison.
Reconstructing all of my piano jazz chord knowledge on the guitar. Painful but still fun.
1
After hitting my 1GB mobile Internet limit three months in a row with one week to go, upgraded to 3GB. #firstworldproblems
RT @runarorama: Going "mainstream" should not be a goal. The status quo is not a stream. It is a stagnant swamp. And we have to blast the d…
RT @peterseibel: "The problem is that coding isn’t fun if all you can do is call things out of a library, if you can’t write the library yo…
RT @mccv: New interview question: "what's your favorite build system?" Correct answers: 1) Rage face 2) None of them 3) Literal table flip
Replying to @pavlobaron
Or you do it just anyway, you know, as a form of premature frameworkization ;)
Replying to @pavlobaron
hehe ;) And then you get an AbstractTwitterSingletonProxyFactoryBeanImpl.
When you've done so much work with raw JSON Twitter data, a lib like Twitter4J (otherwise awesome) feels like working with snow gloves.
1
Muss gerade daran denken, dass auf dem Tisch im Hörsaal Mathematik in Bonn mal "Achs-Xe und Achsilon-Ypse" stand.
@muratk3n Exactly! I see this so often in young people learning ML. Then, as you grow older, well... you know what I'm talking about ;)
@muratk3n Like reeeeeaaaly deeply understands the data, you know ;) It's just bag of words! All the order information is lost!!
@muratk3n hehe, no I meant more like you fit some stuff against a bag of word rep and think the machine understands the data.
Replying to @regisb
@LinkedIn good, I already thought it was just me. Also, CSS missing on first load 50% of the time.
1
That moment in Javascript, when s is a string, but s, s + "", and "" + s all show different behavior.
1
I knew working on timeline summarization is dangerous. It's getting harder and harder to read through hours of backlog tweets.
1
German parliament passes a law that Google News and friends would have to pay to news sites. Google: opt-in for free, or you're out ;)
RT @vivekhaldar: Mindfulness sells - Wired takes a deep look at the rising uptake of meditation and mindfulness in Silicon... http://t.co/3…
RT @jsteeleeditor: Cell phones are like yawning: once someone in a group starts, everyone else will inevitably follow. #fridaynight
I really don't understand the reasons behind the Scala version hell. Do they really break byte code compatability on minor versions?
"A crowded market space only means there's lots of opportunities." Greatest thing I heard yesterday. And so un-German you can't even imagine
1
@pingtimeout yeah, that was the 100th retweet. I wish I could give you a prize, but at least I can thank you ;)
1
Replying to @ChrisDiehl
it was actually ok. I remembered being quite nervous. The talk is also very dense. Could be easily expanded to 3-4 lectures.
Replying to @blattnerma
hehe "You know, if you think about it, it's actually easy to see, right?" ;)
@muratk3n hehe. Yeah, the more nervous I get, the less brain capacity is left to try to hide my German accent ;)
Ok, words I should definitely work on trying to reduce when giving a talk: "right?", "actually", "basically", "easily". ;)
1
As a friend of mine who sits in a fully boarded plane at #TXL just told me, the President seems to have landed a few minuteas ago.
RT @bigdata: Is the sample good enough? Not quite: Twitter firehose vs data obtained through Twitter’s sampled streaming API http://t.co/Yd…
Replying to @alung
@bigdata ... approximating a parametric model with a non-parametric one ;)
Replying to @alung
@bigdata ah, the Kleinberg paper. Which is interesting, because IIRC that's a parametric model. So the post is more about ...
RT @sbtourist: Storm on YARN sounds cool, but I'm not sure having batch and realtime compete for same resources is a good idea: http://t.co…
RT @dominik: Why, oh why, does something in this code reference a file in “c:\projects\“ ?
Replying to @abursuc
also, I think it would be totally wrong to do this as research program. The money would be gone anyway, of course.
Replying to @abursuc
yep. But so far it's just a bunch of politicians talking. I don't think this gets anywhere.
Replying to @purbon
it's completely ridiculous, just some German politicians trying to raise their digital street cred ;)
Replying to @purbon
they want to build it only because of PRISM. Sure we could do better ;)
Replying to @fasihsignal
interesting. I'm always open to improving usage of lapack. Why don't we continue on the jblas-users mailing list. More space ;)
Replying to @fasihsignal
but JNI code is autogenerated. Just needs an entry in the Makefile. If you say you need it, I'll recompile.
Replying to @fasihsignal
sorry, so far I only added what I needed. There are so many of them ;)
It's slightly unsettling that the kitchen magnet doesn't stick to the stainless steel bread basket.
1
Replying to @dominik
yeah, I know. In the end, I copy&pasted them into libreoffice... .
Replying to @fasihsignal
yeah, for things which go through the memory only once, doing it in Java is much faster than through JNI.
RT @mircomusolesi: I have a postdoc opening in network science and big data analysis at the Uni of Birmingham - please spread the word! htt…
Replying to @debasishg
Those seem hard because you need to consider all of your data for every iteration.
Replying to @debasishg
oh, nothing in particular, I was thinking of the kind of graph structure algorithms used, e.g. to identify communities.
Replying to @debasishg
Gregoire Montavon in our group does a lot of learning on deep NNs. But usually, you still use backprop with a few tricks.
Replying to @debasishg
algorithms which work on structured data (e.g. graphs) are a different story.
Replying to @debasishg
don't know all the details. The good thing about conv nets is that they are small enough to fit into memory.
ML methods which you can scale are often either by accident or only become scalable after simplifying the model significantly.
1
ML cares about computation times a bit, but very little about locality/decoupling, the stuff which makes it easy to scale.
1 3
The point is, (classical) ML really doesn't care about computation that much, it's all about modelling.
So I agreed to be a NIPS reviewer this year again. What a clash of cultures between that and Big Data stuff.
@haiqus honestly, the most frustrating discovery was that new math always felt equally hard because the level of abstraction went up as well
Replying to @purbon
@twarko yeah they have to. Kölsch *really* gets stale quickly. Some people refuse to call it beer because ti's too close to water ;)
Replying to @purbon
@twarko Every time I'm back home and order a beer and get a Kölsch I think "ah I'm home".
Replying to @purbon
@twarko hehe, no, it's a piece of home to me. Also, the service has adapted very well. Getting a new beer takes 1min max. ;)
1
RT @fasihsignal: @mikiobraun in Clojure repl, I can (import '[org.jblas FloatMatrix]) but can't (require 'orb.jblas)? I want to play with j…
Conference management systems (looking at you M$ CMT) which log you out after a time of inactivitiy. #allinputlost #rage
Replying to @marcua
@johnmyleswhite oh what a coincidence. Will add more than examples and links!
Replying to @PatriciaHoffman
so far we mostly use heavy hitter algorithms. They let you do range queries which is nice. Will look into DenStream!
The wasted lifetime by grant proposals written in Word without any form of version control. Proposal.doc, Proposal_V1.doc, Proposal_V2.doc..
2
Replying to @mnick
I meant there's a difference between getting infos on specific people vs. we're running larg scale analyses against the whole data.
Oh wait, did PRISM even include large scale data mining? Wasn't that this other story on the NSA and the Verizon data?
1
Or put differently, Big Data = Big Brother. That the NSA used open source Big Data tools didn't really help here...
RT @bradfordcross: There's a perfect storm joke in here somehwere about @Prismatic raising $100MM at a $1BLN valuation and pivoting into cl…
RT @BenDylan: The NSA might be reading all our mail, but at least they use open source :) (openstack, hdfs, cloudbase…)
RT @DRMacIver: It is a source of constant amazement to me how bad Amazon are at recommendations.
Quite positively surprised by the turnout at the #LDNW, even considering it's an event geared at the general public.
Streamdrill, Twitter trends, serienradar, Twitter stocks visualization demos at #LNDW, TU Berlin, EN building, room 181.
Alright, have to cross-check: retweet this if python/ipython is your favorite command line calculator ;)
28 1
Replying to @karpathy
yeah, optimizing some cost functions is just such a powerful approach for almost everything.
Replying to @karpathy
the prof where I did my Ph.D. also did computer vision, so I got some exposure there, too, although I never worked on it myself.
RT @jamesiry: So PRISM collects data from social media, too? Move over Klout, I wanna know my PRISM score.
RT @isaach: ha! RT @samir: It just burns me that the government may have had access to my twitter archives before I did
Advice I got on when to move to the Valley: When you're there, you feel like it's really taking off, grinding to a halt when you're back.
Replying to @syhw
@MyCozyCloud The key here is to make it so easy to use that people are not even aware they have their own VM. More like an iPhone.
Replying to @syhw
I'm convinced that you can even create The Next Social Network based in such a service.
Ok, here's what @Dropbox et al. should do: (1) more API, become app cloud storage (2) allow small server side app snippets (3) app store.
1
Replying to @pablochacin
looks like Berlin beat Barcelona this one time: min 12 max 23. Current condition *sunny*, 16C.
Replying to @ChrisDiehl
or as my father likes to say, you gotta try a lot of things so that some succeed.
Replying to @truemped
@zenogantner events are tuples. It gets slower if they are larger, but for pairs or triplets, you should be fine. So <1k.
Ted's usual are-you-a-Bayesian-play slightly getting out of hand. They're almost gambling for money. ;)
1
Interesting aspect pointed out by @ted_dunning of #recsys systems: exploitation vs. exploration. Add some noise to gain more data.
4 1
Philipp Rösler should have taken a few German VCs as well to Silicon Valley. Would've been great for contacts, culture, and collaborations.
Replying to @munterluggauer
ich glaube beides. Der @thinkberg meint aber, dass man jetzt schon kaum jemand zu einer Festanstellung überreden kann.
Replying to @mpompery
@zeit_geist yeah. Actually, I think the whole purpose was to mention the funding round for researchgate and to call it a future IT beacon.;)
Replying to @fhuszar
Then again, any hype which convinces people to work with Berlin based startups (like us) is good ;)
Replying to @fhuszar
I agree. Also, I think it's also a matter of scale. VCs in the Valley have hundreds of millions under mgmt, in Germany tens.
Seems Lucene is also turning into a general purposes data analysis / ranking engine. #bbuzz
2 1
Replying to @sscdotopen
I agree. How do you see the connection to that kind of iterative map reduce we've talked about?
Rereading my talk's abstract to remind me what I promised to talk about. I actually like it ;)
2
Hm Giraph not only for graph problems but as a general scalable message passing iterative compute engine. Interesting... #bbuzz
1
RT @jakubzavrel: Giraph is cool! Whole graph computations at Facebook are no longer "weekend jobs", but "coffee breaks" #bbuzz
BTW, I'm at #bbuzz for the next two days. Feel feel to temporarily mute me, or chat with me in person.
1
Might well be the largest concentration of Twitter followers/followees for me so far. #bbuzz
Replying to @ChrisDiehl
@twarko hi, @ChrisDiehl suggested we hook up. Are you also at the brrlinbuzzwords?
@moellus wird das die neue Interviewfrage: warum ist das iPad eckig und nicht rund? Ach ne, war ja anders herum...
Replying to @munterluggauer
@berlinbuzzwords @twimpact manches wird wohl aufgenommen. Sonst muss sich der @thinkberg mit seinem iPhone in Reihe 1 setzen
If I'm to believe my student's feedback, I have mad intro writing skills. #plevel #thatstheeasiestpart
Oh great, suddenly no cell phone reception at all. Lighting might have struck. #berlin #summer
When the last tweet is stored, and the last cluster scaled to 10PB, you will realize that you don't want to batch process all that stuff.
Replying to @ChrisDiehl
@thinkberg @gutelius You mean heavy showers every afternoon for the last week would be a very improbable event in California? ;)
Replying to @blattnerma
and I agree, it's a nice property. Helped me simplify formulas quite a few times.
Replying to @blattnerma
It's hard to tell when you have little data whether it's exponential or not ;)
Replying to @5olarpunk
@wirereporter I'm really glad at least one picture (17 of 20) includes a pretty normal open plan work space.
Replying to @alung
that said, I've also worked with professors who only wanted to see the "2-3 most important plots of the paper" first.
Replying to @alung
well a student of mine said he didn't yet see how to structure the paper. So that's what I did for him ;)
Replying to @mdreid
thanks, man ;) I'll pass it along. You know, I'm more in an overseeing sort of function here. Just kidding. A bit.
RT @alung: @mikiobraun I never thought about drawing a storyboard for a scientific paper ;)
I got our #NIPS2013 paper outlined. I wish I wouldn't think about this mostly from a layout point of view. #plevel
Media
1 1
RT @NeelieKroesEU: Blocking & throttling Internet services, apps hurts us all - no reason 2b anti-competitive like this. Pls back me 2 stop…
Replying to @cartazio
@thinkberg yeah, it's still pretty bare and raw right now. But it helps to understand the analysis model.
So Gmail introduces different categories for their emails. Do they also come with different sets of actions for archiving/delete?
Replying to @cartazio
@thinkberg yeah twimpact/streamdrill is also not in there ;) Interesting to see how strong Hadoop is, still.
That moment when you decide to let your kids sleep a little bit longer after the first meeting in the morning was canceled.
Hm, "successful" is probably the wrong word, but he sees the TSA as a role model for execution.
Replying to @purbon
If you want to chat, we can always meet otherwise over a cup of coffee or so ;)
Ah, my blog is down for the moment while we do multiple levels of real and virtual machine upgrades... . Sorry for that.
People in our group asking me for advice on an old method of mine suddenly spikes. Must be closeness to NIPS deadline.
Trying to solve some contract issues with my cable provider over email. #stillbetterthanbeingonhold
Replying to @pablochacin
ah, I see. Well I still write those updates myself. But only if I'm sufficiently annoyed by the weather ;)
Replying to @twiecki
Davon kann in Berlin keine Rede sein, dafür sind die Lebenshaltungskosten natürlich viel niedriger.
Replying to @twiecki
Also uns sagte man, inzwischen seien die Einstiegsgehälter im Valley bei $100k weil es niemanden mehr gibt... .
RT @munterluggauer: @thinkberg wondering how many articles still need to be written to change that notion that an office is solely a place …
Replying to @roidrage
kriegt man das hin, dass jedes mal so ein Modemhandshake ertönt, wenn Daten gesendet werden?
@thinkberg yes. Also, where do I put my books? OTOH, our current offices don't work well, either...
@thinkberg @ChrisDiehl my stock Android client can only put replies into the sent folder, breaking conversion views. Why? They don't care.
@thinkberg @ChrisDiehl it's alread there partly, but requires so much manual setup every time.
@thinkberg @ChrisDiehl doesn't have to be A.I., preconfigured filters to deal with different sorts of email differently would help.
1
It's just incredible how broken email is. It's used for notifications, spam, conversations, data exchange, but clients behave as if it's '93
8 2
And according to the rain radar it's going to stay like this in all of Germany today... #tweather #berlin
Media
There's something fundamentally different between a research project proposal and a startup pitch deck which just drives me crazy. ;)
1
So the main reason people are excited about Google Hangouts is that it's cross platform because of the web client?
RT @berlinbuzzwords: Meet @mikiobraun and @ted_dunning on June 5th at the 8th Recommender Stammtisch at the @plista office http://t.co/2dRC…
Yesterday it rained all day. Today it looks sunny again but it was like 4°C in the morning. #brrr #berlin
Replying to @cartazio
we should definitely meet when I'm in your area. I need to verify first hand ;)
Well, the new flickr definitely looks as if it was built in six weeks... . #yougetwhatyoupayfor
Twitter suggests a couple of German soccer related accounts to follow. Alright, I might be German, but still... .
1
Really funny to see German mainstream media trying to explain tumblr. I always say "tumblr is [was?] the crack cocaine of social media."
1
Replying to @tyldurd
yeah, as some analyst said, it'll be tough to monetize all those page views.
Friend of mine lives in Hannover which is always autocorrected to hangover. Guess we will soon have a new language called Auto-english.
RT @smolix: @mikiobraun @gappy3000 Use the distribution over pairwise distances. Pick the 5% 10% 50% 90% and 95% quantiles as candidates. S…
@gappy3000 @smolix dimension should also play a role. But it's very rough, gives you an idea of the general scale.
@gappy3000 @smolix ah I forgot. Was it even nearest neighbor based? It was something very simple like the diameter of the data / 50 or so.
1
RT @markab: I can't state this enough: Programmers don't burn out on hard work, they burn out on change-with-the-wind directives and not 's…
Replying to @jane_fel_reed
yeah, the Internet used to be about decentralized services, open standards, etc. But now we get mainframes&terminals again.
Replying to @jane_fel_reed
ah, I meant it's just a transport layer. The whole ideas of decentralization, everyone can add a server and so on disappear.
The idea of the Internet is slowly dying. All we'll be left with is a transport layer for data connecting proprietary cloud fortresses.
6
@moellus @baeschtl aber so was von. Was passiert da nur wieder im kalifornischen Bergblick?
Ah nothing like burning through 20% of your battery on your commute and morning coffee. #firstworldproblems
Replying to @DRMacIver
"math problems no one cares about" Isn't that the default? ;) (apart from other mathematicians, hopefully)
First I thought this was a late April fool's joke, but they're killed GTalk and go away from XMPP.
Replying to @padjiman
@MassimoMorelli hehe. Maybe they've started implementing the approximate counters I'm always talking about ;-)
1
RT @AdblockPlus: The request from German publishers to turn off #adblock led to +129% @AdblockPlus installs & +167% donations. Thanks! http…
My follower numbers look pretty eventually consistent lately. Going up and down without notifications... .
Pushed a quick fix to that #jblas pseudo-inverses on Mac OS bug. Pull latest commit if you ran into this.
Transferring blindly at half past midnight, only five minutes to wait for the train to take you home. #thankyouberlin
I feel like I've slowly processed our Silicon Valley trip to a point where I could blog about it. Which I should because it was awesome.
Replying to @tyldurd
totally agree. People who say RSS is dead have no idea how universal it already is.
The call also contains such awesome written-German-only phrases like "völlig neu" (entirely new, you know not just new)
Replying to @purbon
sure ;) but I think a call should also guide research. If everyone just takes the money and does what he always did, that's bad.
They are also including all branches if science which have lots of data. Ok, but then who doesn't.
There's a huge Big Data call out by the German Ministry for Edu&Research which is worded so inclusively everyone is suddenly being big data.
1
RT @ML_Hipster: Just had a confusing conversation with a New Zealand entomologist. I thought we shared a common interest; it turns out he's…
For example, it handles English or German simultaneously. No more switching inbetween texts.
Still miss the physical keyboard, though. But large screen definitely helps. And even autocorrect seems to be a bit better than before.
A week in still super happy with my S3 LTE. Best smartphone I've owned so far. Fast, lots of mem, even like the additiona by Samsung.
1
And 5 minutes of that were spent trying to figure out why the app downloads 1MB of data every day.
Replying to @ChrisDiehl
hehe. It seems I have managed the expectations of my audience well. ;)
Replying to @random_walker
I suspect this is partly necessary to keep people in academia. ;-)
Just want to take a second to point out that it's past 10pm and still about 21ºC outside. #Berlin #finally
1 1
@TanjaAzderska and he's let being in charge because no one else can make decisions as quickly as him ;)
Too often, a machine learning algorithm is just like an uniformed outsider who happens to get it right most of the time.
3 2
RT @drosselkom: Die #rp13 Leute sitzen alle auf einem Haufen und belasten dennoch im Sekundentakt unser Netz. Redet doch miteinander verdam…
@thinkberg @moellus Sowas passiert wenn man so alt ist, dass man Spiele nach dem Gameplay bewertet.
That people will be massively irritated when asked to pay a substantial amount to a company which hasn't added any value - big surprise!
I don't get how anyone could have considered that hackathon trademark idea viable. Maybe besides from a pure business perspective.
Replying to @helpingtheblind
and I don't care about LTE, but that version has 2GB of RAM, not "just" 1GB ;)
Replying to @helpingtheblind
Samsung S3 LTE. Prices nicely dropped now that the S4 is out ;)
Alright. Android this is your last chance. If quad-core and 2GB of RAM don't help, my next phone is an iPhone.
1
Replying to @spidaman
@bigdata @jeremy_carroll ... powerful enough. (Although I might be biased ;))
Replying to @spidaman
@bigdata @jeremy_carroll .. you get more data. Streamdrill does aggregation + resource control on the fly. Not as general, but ..
Replying to @spidaman
@bigdata @jeremy_carroll AFAIK druid is an in-mem columnar store, good for quick scans over data. You'll need to scale out if
1
Replying to @jeremy_carroll
@spidaman @bigdata care to elaborate why? Is it cost or ease of deployment concerns?
Foursquare picking up in Germany after all? Suddenly getting likes for tips I wrote two years ago.
1
Telekom's plans are clear: either the customer using, or the company offering a data intensive service needs to pay to get around this.
2
In case you haven't heard German Telekom is capping bandwidth on DSL. At 16 Mbit/s, you are downgraded to 384kbit/s after 75GB per month.
1
Darn. My Xperia phone is so laggy I'm almost missing calls because the phone app takes ages to load. #firstworldproblems
.@thinkberg: "Now we can make some room on the server" Next up: "I accidently the whole server". ;)
Listening to some weird free jazz vibraphone/trombone duet for the past hour from the Jazz institute across the backyard.
Putting together my stream mining related blog posts into the ultimate streamdrill whitepaper ;)
Replying to @superglaze
they've already been hard at hiring machine learning folks the last 2 months.
Seems our new offices share a backyard with the Jazz institute. Well, at least today it isn't time for tuba practice.
Replying to @purbon
still not over all that snow. :( Always afraid I wake up one morning and it's back ;)
Ich zum Busfahrer "was muss ich mit den neuen Abokarten eigentlich machen?" - "Keine Ahnung." #frictionlessrollout ;)
@muratk3n Guess so, classical acqui-hire. In the best case, they're going to build something similar for Yahoo eventually.
@muratk3n Nope, judging by their blog post seems you get your money back and your data. That's it.
Thanks, I used to have a useful app, now I'm left with a bunch of useless XML files. #astrid $YHOO
1
Incredible, they're just winding down their business: "[W]e’ll be in touch with users shortly to share how to download data." #wererichthx
Replying to @ChrisDiehl
given that he's really into esoterics and stuff, that was pretty unexpected ;)
Replying to @fhuszar
you probably can if you are Google or have the kind of money to go through the whole process.
Replying to @timmoreton
@cloudjunky That's right, but there is something in that picture which people get.
Replying to @mpompery
@zeit_geist @roidrage @thinkberg Ist das so ein Fall von "Nur weil ich Pessimist bin heisst ja nicht, das alles gut wird"? ;)
Apparently, the main reason you need a smartphone, a tablet, and a laptop is that each one comes with its own 5-8 hours of battery time.
Replying to @cloudjunky
maybe the "BEF triangle" doesn't contain the whole picture. Truth is, you can get fast either through scale or approximation.
Replying to @ChrisDiehl
not that it's not a mess here, sometimes, too ;) But a well-connected mess.
It seems to be deeply human, but merely evaluating how X relates to A, B, and C seldom does either element justice.
RT @miss_assmann: Die @drosselkom twittert jetzt auch ;) „Kennen Sie unsere Notifications? Da schicken wir Ihnen ein Fax, sobald ihr Downlo…
Replying to @srisatish
yeah, sorry, we couldn't make it to your offices. Our schedule was pretty much maxed out. Good look with H2O looks very nice!
Winding down at SFO. Traffic crawling along the 101 in the distance, and there's that funny wall of clouds again. #sigh
At Ritual Coffe Roaster on Valencia St. This place can easily compete in terms of Hipster density with Hackescher Markt in Berlin ;)
Replying to @stefan_will
we have to be at the airport at 5pm. Can you make it happen before that? ;)
Another day, another startup. Hanging out at Flurry's reception area waiting for another meeting.
4
Replying to @cloudjunky
thanks! Yeah, it's certainly marketed that way. Saw it first by Acunu's Tim Wilkie. Hadn't given proper credit.
RT @gilgul: When your Twitter friend turns out to be the Boston bomber? Dugg into @J_Tsar's data and here's what we found - http://t.co/MW6…
@muratk3n A few years back, I was in Santa Fe and I walked around a bit. People constantly stopped me to ask whether I was lost.
@muratk3n Never thought I'd miss that kilometer or so I have to walk to the subway in Berlin every day ;)
For example: @thinkberg: "we're going to be the Apple of Data Science" #seriousnotserious
1
Actually driving around the Peninsula is nice because you have time to discuss and evaluate what happened.
1
RT @SFMachineLearn: Note: @mikiobraun talk tonight is *different* from Monday's SF Data Mining meetup (more machine learning focus) http://…
RT @davidandrzej: Looking fwd to meetup talk this evening from @mikiobraun - "Real-time Online Learning for Event Streams" http://t.co/x5Uu…
Great dinner with @gutelius, @ChrisDiehl, and @thinkberg. Let's do it again some time! ;)
1
RT @bigdatasci: "Data Science doesn't have to be rocket science" - @srisatish via SF Data Mining Meetup #H2O
RT @ilparone: "It's not all about scaling" -@mikiobraun SF Data Mining Meetup #bigdata #data http://t.co/PHBGJf15zN
Media
RT @bradfordcross: Yep, big social platforms are ripe to be dislodged. “@andrewchen: Why developers are leaving the Facebook platform http:…
Again at the Paris Baguette in Palo Alto. The rate with which we create new habits is insane.
RT @fhuszar: Data + Coffee = Insight. This week is Big Data Week bigdataweek.com and UK Coffee Week ukcoffeeweek.com
Big Data Week - A global festival for big data professionals
Big Data Week is a unique global platform of interconnected community events focusing on the social, political and technological impact of big data.
bigdataweek.com
UK Coffee Week - Celebrating coffee and where it comes from
UK Coffee Week is a nationwide fundraising campaign for Project Waterfall, bringing clean drinking water to coffee growing communities. Thousands of...
www.ukcoffeeweek.com
@moellus @kommandomutti Ehrlich gesagt war die von H&M, hat aber schon ein paar Jährchen auf dem Buckel #Lieblingsstück
Awesome how a coffee, free Wifi, and SIP let's you phone back to family for 0.14c/min instead of 1.59€/min with regular cell phones rates.
So odd to finally be in the same timezone as all those social network sites which are always sending me notifications.
Replying to @andyrtd
"may I help you?" - "just looking" - "ah, you know awful things can happen. We wouldn't want that, right?" ;D
Programs which just give you a vanilla file selector when choosing which program to open some data with. #firefox
Final preparations for our Bay Area trip. Also coincides with the #bigdataweek. Now if that ain't fitting I don't know what is ;)
I hate recently added distinction between Export and Save in GIMP. I'm fine saving my files in jpeg, thank you!
Tired of scaling? Come to my talk at #bbuzz "Beyond scaling: real-time event analysis with stream mining" in the aptly named Maschinenhaus.
3 3
Nothing like letting you sink to the depth of your tiredness for a short while. #stuffyoulearnasaparent
RT @sinanaral: @seanjtaylor worst part of twitter is we see you tweeting instead of writing;)
RT @seanjtaylor: Worst part about co-authoring in Google docs is @sinanaral saying "I didn't see you write anything new this afternoon."
Weather in #Berlin knows no middle ground. Last week it was like 3°C, now it's 20°C. And it's awesome! ;) #notcomplainingforonce
1 1
Dude, just got my phone back. What an odd experience. It seems life DOES go on even if you're not connected.
BTW, @thinkberg and I and going to SF next week. Tweet if you want to meet us and chat about @twimpact, streamdrill, or jblas.
1 1
Forgot my phone in the office. Periodically checking Google Latitude to see if it's still there. #dontjudge
Replying to @alung
It probably just shows how big the divide between academia and industry is... . Not saying who is leading, though ;)
Switching between tasks has started to take almost as long on my Xperia phone than on my old Nokia E61i. At 5x the clockspeed. #bitrod
Replying to @Xubuntu
ah I see. Can't help you then, I'm perfectly fine with the default L&F. Good job! ;)
Replying to @munterluggauer
@alung we did some solid basic research on image classification. Don't know what came out of the more product oriented WPs.
In retrospect, I was right to think I should have invested in that Taylor travel guitar. If only I had listened to myself.
Replying to @alung
But I think the project was much too big, too many big players involved. Had some nice output, though. But not the EU search engine.
Ok, in summary, it seems like the EU is making a genuine effort in funding Big Data topics, so far, so good ;)
1
Replying to @alung
Yes, I agree. Someone once told me the story why this had to happen, but I forgot. I think this went up pretty high in the hierarchy.
Replying to @vadimkantorov
Nothing wrong with that ;) I was associated with THESEUS, the German spin-off for some time ;)
Replying to @alung
Yeah, we were involved in the THESEUS German spin-off, funded work on image retrieval/classification. ;)
Replying to @alung
It's probably mostly that "Big Data" is an unfunded term in the EU research portfolio so far ;)
RT @alung: .@mikiobraun US : "we lack engineers". EU : "but we thought they were at your place" #bigdata
@timonk I don't know. Probably it's who's leading the innovation. Or they say Big Data = Google. But then who leads besides Google? ;)
Hm. There seems to be some concern in the EU that all Big Data technology is developed within US companies. What about SAP?
RT @Aaronontheweb: It is astonishing how much technical debt one bad developer can create in a short period of time
I took the subway today in Mallorca. Didn't know there's a subway here, eh? Now you do. That's how I'm protecting your tax Euros!
Replying to @tirsen
@thinkberg Ah, now I get it, a real Apple store, not the Gravis-kind-of-but-not-the real-thing-Apple-store! ;)
IMHO rejecting benchmarks because they often fail to capture what you're interested in seems a bit harsh.
Lots of talk about the usefulness of benchmarks in closing discussion between autonomous cognitive systems and ML people.
Replying to @oliverobst
yeah, thanks, I'll try ;) No Pascal3, though. You get at most one extension with the EU...
On my way to the final @PASCALNetwork steering committee meeting. To Mallorca ;) But don't be jealous. Will be staying indoors all day.
RT @ChrisDiehl: “Goddamn, I am so pensive in sepia. Also don’t forget to crop the Starbucks so people think I’m drinking Philz.” http:// ...
Replying to @peteskomoroch
@Prismatic Pretty heavy on Hadoop related infrastructure, don't you think? ;)
Replying to @hrishikeshio
More like "He's more like a system vendor kind of guy" I don't even know what that's supposed to mean. ;)
I'd like to thank @Xubuntu for not succumbing to that flat tile look and feel like everyone else does. #dropshadow Oh and probably Apple ;)
Replying to @leonpalafox
yeah, the DX looked really nice but was disproportionally expensive.
Replying to @leonpalafox
I still print them for reading. Tried the Kindle, but the screen is still too small.
RT @munterluggauer: @mikiobraun back to the roots or what computers do best: accounting/number crunching. ;)
There's probably a special hell for people who just post links to their own blog posts as part of an argument on Twitter.
I must say eBooks from O'Reilly have started to look much better recently. Just make sure to activate "Publisher Font" on your Kindle.
1
RT @sscdotopen: I'm gonna give a talk about the latest developments in #mahout's recommenders at berlin's big data week #recsys http://t ...
Replying to @fhuszar
I should probably rephrase that it's funny that event data can both be used for and against MR ;)
Replying to @fhuszar
oh yes, totally agree, nothing against MR in general. That might have come across incorrectly in my earlier tweet.
Replying to @fhuszar
that whole in-memory database and MR business likes to overuse the term real-time, too ;)
Replying to @fhuszar
also, I've seen quite a few talks of people trying hard to make batch work for real-time... ;)
Replying to @sethjuarez
You're certainly right. I think they believe if the pile is big enough, something useful will be hidden somewhere. ;)
Replying to @fhuszar
There are approaches which analyze the data as it comes in. Even Google developed new, more incremental schemes like Percolator.
Replying to @fhuszar
Smarter schemes quickly get complex. Also, systems like Hadoop have significant latency before a job starts.
Replying to @fhuszar
In a nutshell, MR tends to run through all of your data in iterations, which takes time and duplicates work.
Replying to @fhuszar
well, depends on your application of course, whether you really need real-time or not.
The point being, of course you can crunch three years of logfiles with map reduce. But how do you keep up with the constant inflow of data?
1 1
Reading the Hadoop O'Reilly book. What's funny is that almost all examples are about event data, what I usually use to bash map reduce. ;)
Replying to @jbenno
deswegen ja Daten + Service. Damit die nicht einfach den Stecker ziehen und ich hab nur noch ein großes sinnloses XML file.
Replying to @jbenno
stimmt. Aber soweit will ich gar nicht gehen. Mir würde es schon reichen, wenn die Daten und Service bei mir bleiben.
Been thinking a bit about alternative online social networks. There exist quite a few open approaches, e.g. identi.ca, tent.io, pump.io etc.
Wow, diaspora* is already about 2 1/2 years old. Also didn't notice that one of the founders has committed suicide... .
On Android, you might have half a dozen different versions out there, but at least this forces people to be backward compatible.
So many people I know who have old iPhones complain that most apps have stopped working because they don't work with old iOS versions.
Replying to @munterluggauer
The main problem was that the initial sync with the cloud always failed but also couldn't pick up from where it stopped.
Replying to @munterluggauer
we were discussing backup solutions over lunch. Quite surprising, how buggy that technology is still in 2013. ;)
Replying to @syhw
yeah, if only they'd tear down all those arbitrary region and country restrictions! ;)
Sometimes I wonder if Google really doesn't get why other people might be concerned or if they just don't care.
Discussing the Google book controversy with @thinkberg and his daughter, it becomes clear that Google is the M$ of the next generation.
@muratk3n yeah. There should also be the xubuntu theme pkgs somewhere. Pure xfce4 is, well, very pure.
DST without kids: 'oh no, I have to get up an hour earlier.' With kids: 'Great, let's just pretend it's already 7am, not 6am.'
@muratk3n no need to reinstall, packages are there, select xfce4 as WM in login screen to give it a try.
@muratk3n if you don't like GNOME 3, xubuntu is also an interesting alternative. Or try the xfce4 pkgs (+ xubuntu themes) in std Ubuntu.
Replying to @ChrisDiehl
also still no sing of spring. Forecasts for next 7 days show no sign of improvement... .
Replying to @ChrisDiehl
I can already see the startups coming who trade in food derivatives based on my sentiment score ;)
RT @nanexllc: #HFT is walking 1/2 block up to steal a cab. Exploratory trading is sending someone else and seeing if the crowd lets him live
Why does the G+ Android app has about 1MB of data per day? Even if I seldom use it? And it still loads an awful amount of stuff if used?
Replying to @blackfemcoders
... but there are more principled ways of dealing with that than most NoSQL DBs attempt.
Replying to @blackfemcoders
... consistency like most NoSQL DBs do is an oversimplifcation. It's true, during failures you have to sacrifice A or C, ...
Replying to @blackfemcoders
thanks for your comments. Although I didn't mean to say CAP is false. Only that using it as an excuse for eventual ...
Is @thinkberg reading RSS feed subscriptions of our friends on ttRSS to me the future of decentralized social networks? ;) #nothingtohide
"Google Keep is another example where they give you some value in exchange for free data to improve personalized search." (by @thinkberg)
1
Replying to @tirsen
After reading that article, it seems that the NAO is mostly a winter phenomenon :(
Trying out Tiny Tiny RSS. Like what I see so far. Even has sharing (via RSS feeds), plugins for Twitter and G+. A nice little monster in php
2
Oh and next time Berlin plans to relocate to Finland, I'd like to know six months in advance.
2 1
RT @dmi: Linux CLI tip of the day? alias ffs=sudo Usage: $ somecommand somecommand: permission denied (not root) $ ffs !! somecommand s ...
RT @sscdotopen: @mikiobraun that's the reason youtube won against google video, as they could iterate faster because people would forgiv ...
Google is trying to stay agile by trying out and axing services, but they've grown too big and people expect stability.
2 1
The problem is if Reader were some startup which ran out of funding people would accept that. But it's backed by Google, so we expect more.
2
Backlash against Google's latest service cuts points to deeper problems with the free services model.
RT @alung: @mikiobraun one of today's talk (about EP-ABC) : "we have no idea why it works, but it does". We'll just have to wait a few y ...
@thinkberg hoffe das liegt nicht nur daran, dass keiner da war der dauernd ins Ky wollte ;)
New Post: Misconceptions about the CAP theorem blog.mikiobraun.de/2013/03/miscon… My thoughts on this good article by Eric Brewer infoq.com/articles/cap-t…
Misconceptions about the CAP Theorem
If you've ever listened to a NoSQL talk, you've probably come across the CAP theorem. The argument usually goes like this: * Traditional databases...
blog.mikiobraun.de
CAP Twelve Years Later: How the "Rules" Have Changed
The CAP theorem asserts that any networked shared-data system can have only two of three desirable properties (Consistency, Availability and...
www.infoq.com
2
What, delicious removed its stacks...? Like half a year ago? And I only realize it now? #notgoodfordelicious
Newspapers say we have good chances to beat the record for latest day in the year with temps below 0C: March 24, 1922.
@muratk3n ah read the vertical data scientists are fake post. In that case I'm multi-layer horizontal, of course ;)
RT @cssquirrel: How to social API: 1. Make API. 2. Invite developers to make cool stuff. 3. Steal ideas. 4. Sunset/restrict API. 5. Kick ...
RT @sgourley: Zadie smith takes on Zuckerburg and our obsession with social networking -- two years old and still hits hard as ever http ...
RT @DRMacIver: Actually The Old Reader looks quite nice. I can't remember what I disliked about it before.
Replying to @syhw
@thinkberg ah I see. Actually I don't care so much how it's technically done as long as the users control the enterprise.
Replying to @syhw
@thinkberg @infinitdotio ah, I was talking about the economic model. Like crowdfunding, but you own a piece of the company.
1
Replying to @mpompery
@zeit_geist I'm more worried about the legal implications of creating a global cooperative. ;)
Actually, social network sites should be run as a form of digital co-op to make sure the users own their data. (talking with @thinkberg)
RT @grumpygamer: I subscribe to close to 500 RSS feeds, I need to be able to scan very very quicky, not look at pretty pictures.
Replying to @mpompery
@zeit_geist we should definitely have another cup of coffee when you've settled in!
Replying to @munterluggauer
I think the outcome of a good redesign depends a lot on whether you've understood what's important. Google's focus is off.
Replying to @munterluggauer
And to me it seems like Google fails to understand that people don't like to change.
Replying to @munterluggauer
yep, that is right. But refactoring can also lead to changes. Which is ok if the other side are other people within Google.
First they get rid of smartphones with physical keyboards, now Google Reader. I'm feeling old ;)
1
I think Google Reader is yet another example how Relentless Refactoring works well for technology but not for end user products.
Recently, lots of people are sending me connect requests on linkedin as if it were Twitter. What's happening?
Replying to @abellogin
ah, no. We did this a few years ago when last.fm's API was much more open and used the info people provided in their profiles.
Replying to @stefan_will
was really what you would expect, girl bands/female singer-songwriters for women, metal for men.
Replying to @stefan_will
yeah, it was quite interesting. We could also identify music tags which are most indicative for male or female listeners.
Replying to @noelwelsh
yeah ;) not that we didn't try. But ML-wise methods were too simple. And we probably didn't try the right journals.
That Facebook PNAS paper reminds me of unpublished work where we predicted gender and age from people's last.fm plays. Worked quite well.
Not so many surprises here, forecasting, fraud detection, price setting identified as some important applications for ML at Amazon.
1
Slideshow laptop's battery died during Amazon's reception. That's not the ops level I'm used to. ;)
1
Replying to @gutelius
@ChrisDiehl well at least proponents of HFT like to say that nothing can go wrong ;)
@ChrisDiehl I think economics people have strong bias that all market related activity will reach equilibrium eventually.
Does algorithmic trading destabilize the market? Can a bunch of distributed agents interact in unforseen ways? Hell, yes!
1 1
Replying to @superglaze
Given the current weather, the U-Bahn might be the best way to get around town ;)
@nanexllc which described all kinds of exploratory or even predatory HFT strategies. Or is the first time it was confirmed on actual data?
Replying to @nanexllc
Nice summary. Also a nice paper. But is it really that surprising? I remember reading a dissertation about algorithmic trading..
Replying to @alung
They're pretty vague, seems to be about ML for the Global Inventory Platform, so maybe inventory opt via sales prediction and stuff;)
People who set tight deadlines but then only reply with cryptic one-liners "sent from my iPhone" when you need their support.
1
Replying to @alung
but seems so far that they're running the whole operation (hiring, organization) from overseas. Phone interviews in the evenings, etc
Replying to @alung
AFAIK, Bezos wanted to get a foot into ML research to rival MSR. Probably didn't care where. Ralf Herbrich is from Berlin, so... .
1
Main reason: it's uncool to be on the same social network as your parent. Just as @thinkberg has been predicting for years now.
Replying to @ChrisDiehl
Ah, I'll be ok. Humans are easy to forget. In two months I'll be complaining about the heat. IF it heats up by then ;)
Replying to @ChrisDiehl
I'm currently in a state of mind where I'm not sure a few months of summer will be enough to make up for this winter.
RT @zaxtax: We need more content like this "Collapsed Gibbs Sampling for Dirichlet Process Gaussian Mixture Models http://t.co/uDEuAnt8v ...
Alright, one last check whether people over in the new world are awake already, and then it's time to publish my Megastore and Spanner post.
Replying to @superglaze
was just talking with @thinkberg over lunch whatever became of those guys ;)
Now having a look at the "Exploratory Trading" paper by Clark-Joseph who analyzed HFT data to uncover trading strategies (via @nanexllc).
3
I'll spare you the obligatory winter and snow pics. Believe me, it's -5C and it ain't pretty (after 4 months of winter, at least).
Replying to @nanexllc
@optionmonster that's your interpretation of what's going on there? The wording of the article is pretty neutral.
Looks like I'll be attending Amazon's reception Tuesday night on their plans for the new research lab in Berlin.
@muratk3n no kidding. I even looked up how to hack the registry under Windows to do the remapping there. Somewhere on my blog...
Replying to @noelwelsh
I agree. Once you start looking you get all kinds of ideas. BTW, streamdrill is with two "l"s.
Och ne, der Hachtmut? Was kann der eigentlich nicht? Hoffentlich bauen die nur mehr Fahrstuhle ein als am Hbf.
Replying to @DRMacIver
Ah, just thought BRAAAAINS is the appropriate thing to say for a follower zombie ;)
Replying to @DRMacIver
Somtimes you have to build the house to see if it'll stand. Luckily, no one got hurt ;)
Remember when you could read about an algorithm in the morning and try it out in the afternoon and not everything was a friggin framework?
5 1
Replying to @DRMacIver
yet another case of the sunk cost fallacy we talked about years ago, I guess ;)
Replying to @DRMacIver
you'll be ok ;) As it is, software is a complex thing and it's just hard to step back and say "we're doing all wrong".
RT @lawrennd: Three cheers for PLoS! Require alphanumeric passwords that expire. But then email it in plain text! Paid for by inflated o ...
Replying to @roidrage
I wish I could, but my 3yo son is so hyperactive sometimes that he constantly slides down from his chair otherwiese.
Told a Ph.D. student some research project war stories. Afterwards he asked: "Ok, so what would you say was your most successful project?"
Admittedly, my phone may be underspec'd for Android 4.1, but still. Google can scale but can they do small? #Glass
Made my Xperia mini pro 500% more responsive by doing a factory reset and reinstalling everything. Something's not right.
Still seeing a lot of winter coats. As if people are reluctant to believe in spring. Well, me too ;)
Replying to @leonpalafox
yeah, we use a lot of neurophysiologically motivated preprocessing and then do LDA on the features. More complex methods have..
Replying to @pranav_modi
hehe. Buying of course. Not sure anyone can actually _read_ the whole thing ;)
Replying to @derMicha
also von deinem Foto konnte ich ihn ganz gut auf'm Bildschirm einscannen ;)
Ah, Germans, going from "it's so dark outside" to "it's so bright outside" in less than 48 hours ;)
2
Replying to @leonpalafox
How was it? In fact, I worked a bit on cognitive workload detection via BCI. Cars and EEG caps and all ;)
RT @janl: “…we could have a phone call tomorrow morning at 09:30…” — “Neither 09:30 nor phone calls exist in my world.”
Replying to @ChrisDiehl
yeah, also: who are they to decide when I can watch some show? Pft. Makes no sense at all ;)
Replying to @ChrisDiehl
yeah. They're like "we want to see it now!" - "You have to wait another hour!" - "Nooo! Why??" The truth is I don't know.
My kids who have been mostly exposed to DVDs and video-on-demand (better controlability) have a really hard time understanding broadcast TV.
1 2
Replying to @superglaze
I think that article is also an excellent example on how to freely misinterpret other sources to fit ones own needs.
Ah I know, replication across data centers via quantum entanglement. Google will figure out a way.
It seems you need a few rack-mount atomic clocks per data center for Google Spanner, though. Wonder what's next. Miniature black holes?
Reading Google's Spanner paper the last few days. Probably more Data Base than Data Science, but still interesting.
2
Replying to @Nico
heute Nachmittag soll endlich mal die Sonne rauskommen, sonst alles ruhig ;)
Replying to @purbon
I think the newspapers reported that we got exceptionally few sunshine hours this year. Spring time!
RT @tirsen: @eventsofa @mikiobraun @thinkberg @roidrage I think a ray or two actually burst out over here for a second! Hurt my eyes.
Replying to @tirsen
@thinkberg @roidrage @eventsofa Yeah, I agree. Kinda makes you start hoping this winter will eventually end ;)
Putting together the last numbers for a project's annual accounting. Remind me why this is my responsibility.
Somehow, the "Data exhaust" image doesn't fit. It's not like you'd have to get rid of the data, just delete the logs.
Me: "What Yahoo cancels home office?" @thinkberg: "You haven't heard? That's old, came in this morning at 6" #internetage
At least Paxos is at the heart of Google's distributed databases Megastore and Spanner to realize transactions and consistency.
Replying to @sscdotopen
ah, then you didn't miss the grand finale ;) snow storms and everything...
Replying to @sscdotopen
when did you arrive? Today it's already back to normal, yesterday was tough. ;)
.@ds_ldn recommended Google's Spanner, reading it right now. Megastore also looks interesting. Probably in a follow-up post ;)
Kann man in #Mitte jetzt nicht mal mehr auf einen Termin warten ohne von der Polizei angesprochen zu werden?
1
Replying to @ogrisel
Put differently, it only helps with the parallelization, it's not a running system you can dynamically query.
I probably need to start putting more of my thoughts into blog posts than into tweets. Then it's probably clearer what I mean. ;)
Replying to @ogrisel
query results. What they all do is dumping results at fixed intervals. Depending on your application, that might be a lot of data
Replying to @ogrisel
No, what I meant was that stream processing frameworks like Storm are all about the parallelization, but there is no support to
@moellus Und ich dachte, du hättest das mit der Alternativrechtschreibung jetzt endgültig auf die Spitze getrieben... ;)
Replying to @fhuszar
yeah, that, too ;) I'm also thinking it would/should be illegal to actually wear them while driving.
Replying to @fhuszar
and since the Glass seems to be pretty small, I'm a bit concerned they get it right this time, that's all ;)
Replying to @fhuszar
IMHO they've built some stuff into Android to make it work on smaller devs, but in the end it doesn't really work well.
Replying to @fhuszar
what device do you have? I think it's ok if it's spec'ed well. But on my 2yo Sony Ericsson, it's very laggy all over.
Replying to @fhuszar
.@fhuszar #ifihadglass I'd still be manually clearing app caches to free memory.
So you're probably as fed up with (my tweets about) the #Berlin weather but skies are clearing up tonight so it'll be icy tomorrow.
Replying to @superglaze
I'm sure there are some consultants out there who claim they can make sure a video goes viral ;)
Replying to @johnmyleswhite
@ogrisel @zenogantner @IgorCarron BTW, what's the limit for number of @-replies in a tweet? ;)
Replying to @johnmyleswhite
@ogrisel @zenogantner @IgorCarron ... and the results might not be useful, but hey..."
Replying to @johnmyleswhite
@ogrisel @zenogantner @IgorCarron ... boost your profile has a high tec company. Your customers might sue you for that ...
Replying to @johnmyleswhite
@ogrisel @zenogantner @IgorCarron Yes ;) Still, nothing I'd recommend to a company. "Just do a public competition, to ...
Replying to @ogrisel
@zenogantner @IgorCarron But for netflix, it wasn't. The winning method was too complex, and they got into legal troubles.
Replying to @ogrisel
@zenogantner @IgorCarron Probably I'm not explaining myself well ;) We're all happy we have that data, of course.
@zenogantner @IgorCarron So yeah, it was interesting and worthwhile for research, but for the company, it wasn't a success.
@zenogantner @IgorCarron True, but at considerable legal risk for netflix. In fact, it was so high that they didn't repeat the experience.
Replying to @alansaid
yeah, I meant it wasn't a success as a model for exchanging data between industry and academia.
@zenogantner @IgorCarron but with real data from real people, things can get pretty complicated pretty quickly.
@zenogantner @IgorCarron Often when discussing projects, people say "and then we make something like that netflix challenge".
Replying to @tyldurd
Yeah, also had that thought. If all that snow would have been rain... *shudder*
@muratk3n @god Supposedly, snow's gonna stop in the next few days, then there will be sun and eventually temperatures above 10°C.
So far, this week was PostDoc Mgmt Week: Monday project report, Tuesday job interview, today project accounting. Lot's of AC, at least.
Replying to @debasishg
I'm always suspicious why the math becomes so complicated mid-computation. Maybe there's a short-cut. ;)
Replying to @seanjtaylor
it's like marketing isn't even trying to highlight actual features anymore.
Replying to @superglaze
yeah, most people are surprised. We did some language analysis on Twitter, about 30% of all retweets are from Indonesia.
Replying to @ChrisDiehl
tried to take pictures but the wind was so strong you couldn't see the snowflakes. It's a bit above 0C so it melts fast at least
Replying to @sscdotopen
I know that view. The amazing thing is that the hills in the far distance are already part of Lebanon.
Replying to @debasishg
I have to admit I'm also struggling with all the details. I'm more a frequentist type of guy ;)
Replying to @debasishg
The book by Chris Bishop has some background. Computations easily become very messy, though. But the end result is often simple.
1
Replying to @debasishg
I guess depending on who you ask, both methods work fine. Seems to depend on what's in your toolbox.
Replying to @debasishg
sampling or variational approximation. So saying that it's intractable has minor practical implications. I might be wrong, though
1
Replying to @debasishg
most complex Bayesian models can't be used for inference in an exact fashion. Usually one uses some form of approximation, either
1
Replying to @debasishg
which would matter if there was a single Bayesian method doing exact inference - apart from conjugate priors. ;)
1
Replying to @tirsen
if it helps, it's the wet kind which will probably have thawed by tonight. A wait, that doesn't really help. #theworstkind ;)
Option --post-doc: suppress all details on error messages, because the PostDoc will pass it along to his grad students anyway.
1 1
Also, one of these days I'll hunt down the code for computing the free memory and ask the responsible coders WTF they were thinking.
Android's memeory management partition into internal/SD-card is a clear case of premature optimization.
People with good eyesight have no idea how hard it can be to locate your glasses if you've put them in a non-standard location.
3
Replying to @dekdev
Jo. Ich bin ja auch kein GEMAjünger, aber mir geht Google's scheinheiliges Verhalten auf den Senkel.
Replying to @dekdev
Ich glaube, die GEMA hat da gar nichts gegen, deren Auftrag ist "nur", den Anteil für die Künstler einzusammeln.
Replying to @dekdev
Davon profitiert im Endeffekt nur Google, weil die dann sagen können "wir müssen nicht zahlen, in DE sieht man die Videos ja nicht."
Replying to @dekdev
ah, das ist ein Missverständnis. YouTube sperrt die Videos selbst präventiv, die GEMA hat damit aktiv nichts zu tun.
@muratk3n On first sight looks like a stream processing framework like Storm or Yahoo's S4 atop Hadoop infrastructure.
Replying to @gema_news
@GEMAdialog Himmel, ihr habt heute morgen aber einiges zu tun. Ich dachte auch, es gäbe eine einstweilige Verfügung gegen diese Sperrtafel.
Replying to @KESS_
außerdem ist es irreführend, dass YouTube behauptet, die GEMA hätte die Rechte nicht eingeräumt, YouTube sind die Kosten nur zu hoch.
Replying to @KESS_
Bin kein Gemajünger, aber das macht YouTube (bzw. Google) schon präventiv selbst... .
Reading Twitter this morning I get flashbacks of Liv Tyler and Ben Affleck listening to Aerosmith.
Like every year around Feb it feels as if it's never going to be warm again. Around 0C forever. Also haven't seen the sun for weeks now.
Time to earn me some more Academic Coins, bunch of reviews and other admin stuff coming up... .
Replying to @nglance
quite an obsession with sweaty armpits lately... ;) has your account been hacked?
Replying to @superglaze
The worst about the QWERTZ keyboard is that they've moved all the brackets to AltGr+... . Still I'm stuck with it... . ;)
Well, let's see how well Twitter's new predicted language features handles indonesian and mixed language tweets... .
Started playing Final Fantasy XIII yesterday evening. It is the very definition of Uncanny Valley. And CG hair.
Replying to @SoundCloudJobs
@SoundCloud Either me or @thinkberg, we counted to three and clicked ;)
Apparently they forgot the wireless network in our new building. So now they're drilling holes for the last month. #soundsreasonable
@gappy3000 @jeffbercovici Oh yeah, forgot about those. I think the case of Lehrer is peculiar. (BTW, "Lehrer" is German for "teacher")
Replying to @syhw
no problem. I kinda snuck into a PostDoc myself, too ;) But in retrospect, I wasn't ready to leave back then.
Replying to @jakubzavrel
I agree. Ph.D. is perfectly fine, but doing a PostDoc is quite another thing.
Replying to @syhw
other than that, there is less freedom in research once you have to hunt for grants, you should be ready for that if you do a PostDoc.
Replying to @syhw
... how well you can plan something. Which is naturally a problem if you take too much of an academic approach.
Replying to @syhw
... others, while it's true that there is a stronger focus on stuff that works, there are other constraints as well, for example ...
Replying to @syhw
In particular, concerning lack of fraction and doing interesting and relevant stuff. But from my experience and conversations with ...
Replying to @syhw
well, being somewhat disenchanted with academia, they now believe that everything's better in the industry.
Listening to the recent Ph.D.s in our lab contemplate academia vs. industry issues, I feel old. #theystillhavesomuchtolearn
Random walking through the 1.5k song on my phone for the past few days. How hard can it be to get slightly more sensible random playlists?
Replying to @esammer
oh, I also didn't want to suggest that Swing provided a native UI feeling BTW... ;) Far from it...
Replying to @_krisjack
It's not that we don't get paid, only that the pay is so oddly uncorrelated (= constant) from what you do.
Still unclear where to rate a paper. 1000 AC? More? (If you should average about 2000 AC /month)
So far, I'm up to about 40 AC for today ;) I guess 100 AC per day (so about 2000 AC) per month should be ok.
Replying to @alung
Exactly ;) Later installments of the series include "PostDoc", and then "Professor".
Replying to @alung
And the occasionaly 'Achievement Unlocked' badge for your first published paper, first project etc.
Replying to @alung
No, just the warm feeling that you've actually accomplished something ;) And you can brag about it with your peers, of course ;)
For example: attending a talk = 10AC, meeting with a student = 20AC, reviewing a paper = 50-100AC (depending on length).
3 1
Contemplating the introduction of Academic Credits - some form of virtual currency to measure all that unpaid and unaccounted work.
5
Attending Ph.D. defense, unfortunately procedure calls for up to 90 min. of discussions after the 30 min talk. #OXYGEN!!
Replying to @esammer
remind me again how eclipse has super native feeling UI because they're not using Swing ;)
RT @twimpact: If you want to know what's going on at Vatikan-city. We capture the Pope tweets and pipe them into streamdrill.... http:// ...
RT @twimpact: The German SPIEGEL Online published it last week already, but today it will be in the paper. We have helped to... http://t ...
Replying to @xyzdkfe
@omitevski yeah ;) But I fear we first go through a phase when everyone claims to be one. It'll take time to figure out what we're worth ;)
@erich_owens "data scientist" is slowly picking up, but slowly. Heck, even "Big Data" has just been picked up by newspapers.
Replying to @xyzdkfe
@omitevski yeah. But I wonder how many industries there are in Germany where your pay scales massively with impact besides finance.
Replying to @xyzdkfe
@omitevski I have no numbers, but I doubt it. Germany's much too conservative to get crazy about data science that quickly.
Verstehe ich das richtig? Rücktritt wegen der Komplikationen, wenn die Bildungsministerin eine Hochschule verklagt?
Also "Flüchtlichkeitsfehler" suggeriert ja, dass mal eben so aus Unachtsamkeit was passiert ist. Irgendwie passt das nicht.
Replying to @peter_c_william
.@peter_c_william interest was probably the wrong word. It's that 'this is so cool, everybody needs to know'-feeling.
Sometimes I postpone a blog post to let my thoughts ripe a bit more. Instead I just loose interest in the topic.
It's odd how you get used to doing stuff for free (reviews, organization, editorial work) for strangers when working in academia.
3 1
Oh great and now my disk is full because of all those hosted OSes... . #500GBusedtobealot #goodoldays #IcouldfitanOSon1.44MB
Replying to @superglaze
Let me guess, you can only get them prepaid in packages of at least 500 points?
Oh great, seems my xubuntu distro's libc is too recent to run #jblas on centos. Luckily we have virtualization technology.
RT @oising: Dev: "It works on my machine, just not on the server." Me: "Ok, backup your mail. We're putting your laptop into production."
Catching up on administrative stuff after a 2d OOO while contemplating whether to post about PAC-Bayesian bounds. #choices
@_nipra seems to be based on jfreechart. Wasn't that impressed with that when I tried it out.
@moellus ein Becher Joghurt, mehr Zeit bleibt beim Bedienen^H^H^H meiner zwei Kindern nicht. #fryhstyck #hierso
Replying to @wrede
only install the JDK, copying all plugins and setting paths by hand is the only way around this mess. /cc @Oracle
Last night I dreamt I saw one of the first 4K videos on youtube. What a waste of bandwidth, both dreamwise and networkwise!
1
I just love how sometimes I'm like "no, you misunderstood me, that won't work" and in that instance an idea pops into my mind how to do it.
It's so much nicer to be outside after temps rose by 15C over the last couple of days. Now if only the rain stopped. #nothingbutcomplaints
And I'm also going to remove support for non-SSE3 CPUs in jblas. Doesn't even Atoms have SSE3 nowadays?
This time, I'm also keeping the ATLAS libraries some place safe so I don't have to go through this again.
Rebuild ATLAS for Linux, Mac, and Win the last few days. Let it sit for a day, then I'll bump a point release of jblas with pinv and QR.
Can't people just stop to fiddle around with their software forever, breaking old dependencies? ;) #probablynot
Replying to @syhw
I don't think it'll work. CUBLAS has a different memory management (on the GPU, of course), and a cuda* prefix for the functions.
Replying to @syhw
@thinkberg Looked at, but it's not a 100% drop-in BLAS replacement, is it?
After ATLAS optimized itself to death for 5h on @thinkberg MBP, jblas gets 21.6 GFLOPS @ 2.6GHz for matrix-matrix-mult. #wellworththewait
Letting @thinkberg build ATLAS on his MBP, getting occasional feedback of the most annoying build messages via chat.
Replying to @huitseeker
@thinkberg ;) I don't think we have something where we talk a lot about the technical details of it all.
Replying to @janl
@zef What's most unsettling is that there are a lot of people who have (accidentally?) committed their whole $HOME to github.
1
Replying to @huitseeker
@syhw Oh yeah ;) When we started, I was into Ruby, and @thinkberg used Groovy, so we figured we needed some new common base. ;)
Replying to @agnoster
@syhw Yeah, but my point was that Maven is ok once you get used to it, too ;)
Replying to @syhw
Yeah, Scala is good, doing all my work for the last 2-3 years in it, but hard to beat level of maturity for maven support in IDEs.
"Is Scala hard?" - "No, I'll show you how to set up a Maven env in no time" - "Maven? Ah, forget about it" #everyfriggintime
Gaaah, found a glaring bug in jblas's JNI code. Missing return statement. What is this? 1997? And how could that code still work?
RT @robinwauters: Know what this is? It's the copying machine at Rocket Internet HQ :) get it? GET IT? Cc @Rocket_HR http://t.co/4putP7qD
Replying to @alung
I mostly doubt that you can generalize well (e.g. across topics) based on so little data.
Replying to @alung
I did that sort of batch hand-labelling for tweet languages once. I think you can do 500 or so in an hour.
What, wait, that Sentiment Analysis for Twitter competition will be based on about 10k tweets. TEN THOUSAND?
1
@bastianventhur almost typed that, but then realized that some people might say that non-negative (X >= 0) is not equal to positive (X > 0).
Next time you're tempted to write "due to the nonnegativity property of X" just say "because X is not negative".
Some tasks seem to require staring at the work before you for an extended period of time till you accept the inevitability of it all.
RT @ChrisDiehl: Big Data 2.0: Turning the insight knob to 11 AND making you a perfect espresso every time. Makes the insights go down al ...
Replying to @markhuberty
@ChrisDiehl 2.0 will be better than just plain Big Data in practically any way you can imagine! ;)
Replying to @sml8bigdata
You post interesting stuff, but what about your hompage? It seems broken?
@samueljohn_de ah, probably not. More like Big Data 2.0 will be an important step towards "true K.I" ;)
The gorilla arm equivalent of location based services: you're usually indoors when you check in so reception is bad, and its no fun.
Quantum computing will probably help with Big Computing, but not Big Data. (Thanks to @thinkberg for the discussion)
Temps from -8C to -5C and lots of new snow in #Berlin. At these temps the question becomes how long you're going to be exposed to the cold.
1
RT @Elie__: 1st pitch: Becoacht, the Airbnb for coaches. #hyberlin Lots of startups being "the Airbnb for xxx" ^^
Ok, to be exact, they also have a saxophone player and a percussion guy. #notallbrassthatis
On a related note, finally discovered that if I wear the two pairs of handgloves I have over one another, my hands stay warm even at -7C.
Had to add handling strings to my magic Ruby fortran-wrapper scripts. Haven't done Ruby in over two years, I guess.
Adding least squares and pseudo-inverse to jblas proved to be a bit more involved than expected. #onelinefix
@bastianventhur @thinkberg It'd probably help if he didn't share all that stuff directly to Facebook. ;)
Also uninstalled foursquare for good from my phone. And it feels good. #unpaiddataentryservantnomore
Replying to @sscdotopen
aha. Time to have coffee together, it seems ;) Can you send me a link to the paper?
Hey, @amazon, wouldn't it be cool if the Kindle lock screen displayed the cover of the book I'm reading right now?
That #hyberlin startup event is quite a mess information wise... . Dates and times vary quite a lot between website, FB, tickets, ... :(
@robtibshirani you sent me a direct msg with a link to a suspicious looking *.ru site. Could your account have been hacked?
RT @cartazio: @mikiobraun relational algebra is the king of DSL successes, anything nearly as wide spread is unrealistic :)
Replying to @cartazio
yeah, I remember taking a class on that. I was really impressed how elegantly everything fit together.
Replying to @cartazio
I fully agree. It worked quiet well for the RDBMS world, but for general data analysis, it's still a long way to go, IMHO...
@zeit_geist kennste das eben erwähnte whitepaper? Das sollte Dir doch entgegen kommen wg. "big data primitives".
Ah sorry, I meant #darksocial ;) A quick search for #darknet reveals that that is unfortunately related to something entirely different...
Subscribing to a few RSS feeds in Google Reader which @thinkberg sends me over chat like it's 1999... .
Funny that I almost exclusively use the winking ;) instead of :) although I never wink in person.
Anyway, we'd love to hear your comments, rants, anything on streamdrill. We'd like to learn what you'd like to see most in the 1.0 release.
Something's wrong with my eyes. Noticed the color fringes of subpixel smoothing. That's what you get from a weekend away from the screen.
More funny German words: "rumeiern" (v), literally "egg around/roll around like an eg", be indecisive, to vacillate. Past: "rumgeeiert"
1
Replying to @mcfunley
I also liked your argument that there is noise in the data. Exactness suggests precision, not always true.
Uninstalled a few apps and rebooted my phone and now it's super fast. And I'll never know why exactly. #onlychangeonethingatatime
@moellus hehe. Ja, von 2.10 hatte mir der @thinkberg auch erzählt. Wir warten erst mal, bis die andere upgedatet haben. Das dauert immer was
Replying to @alung
yes, forgot to say, we then filter out the German ones those libs & some homebrewn SVMs and bloomfilters.
Replying to @alung
TV related tweets in french? We filter for show related keywords that's about 5M per day for all languages. Don't know how many DE.
Replying to @alung
Germans and Twitter, a long story ;) You think there's more TV related chatter on Twitter in France?
Replying to @DRMacIver
yeah, not unless they can tell me which formal definition of causation they're using ;)
Replying to @alung
yep. However, it's based on the free subsample you can get from Twitter, so real numbers might be higher.
RT @the_real_jambi: #Berlin wins over #sanfrancisco in the great @versus_io comparison engine list. http://t.co/vcb4u4uR
RT @zeit_geist: @mikiobraun simple answer: a copycat is a startup from Germany. therefore time is an-irrelevant
Replying to @mpompery
@zeit_geist Dang, it seems I *really* don't have a post on serienradar yet ;)
Replying to @quantisan
yeah. I guess they'll be spending millions of $$$ on that. Hopefully they'll do something sensible ;)
Replying to @cartazio
Also, how am I supposed to do that as an enduser on an unrooted phone? To me, this looks pretty broken :(
Replying to @cartazio
AFAIK, that space is organized like a filesystem, so memory fragmentation shouldn't be a problem.
Replying to @cartazio
ICS. I suspect that the numbers Android reports are plain useless or at best misleading.
Replying to @cartazio
Downloading works, fine. In fact, it downloads the full 10MB each time but then fails saying it doesn't have enough space.
I'll never understand why Android refuses to install a 10MB app when it claims to have 100MB free app memory.
So let's boot up my Win 7 VM and see whether he'd like to install some updates. #firstthingsfirst #2013
RT @roidrage: Any sufficiently advanced, home-baked monitoring solution is indistinguishable from stream processing.