Mikio L. Braun

ML and AI expert

@mikiobraun

@mikiobraun Twitter Memorial

19,827 tweets · 2008–2024 · 1046 threads

2017

My kids just showed me that Google search can identify songs... I‘m feeling old that I didn‘t knew... but kinda confident that I know roughly how it works ;)
1 14 View thread →
@LauritaApplez I dunno, either time will tell or maybe give it a slight tap to see if we can get it moving again ;)
@xaprb I have similar feelings about Huey Lewis and the News. Apart from sound, there is also some element of joyful irony (I think) that has since been lost in pop music.
View thread →
Replying to @agibsonccc
@Springcoil Yeah, I guess IBM‘s business plans are also grander than your average AI startup...
1
Replying to @agibsonccc
@Springcoil To be honest, I don‘t even know what Watson exactly does. Wasn‘t it more expert system stuff? But yeah, at the rate the hype and misinformation of the public goes, the next AI winter is not improbable.
1
@Infinite_Monkey @agibsonccc You shouldn‘t have to reinvent a field to silence the critics ;) If that‘s what you want and you can find a way to make a living, why the heck not I say...
1
Replying to @Springcoil
@agibsonccc Whaaat? 😂 I think the point is, you shouldn‘t listen to people who have no idea what they‘re talking about ;)
Replying to @agibsonccc
In 1994, when I was considering studying CS, people questioned this because they said that computers can program themselves soon...
1 3
Replying to @stadtlegende
Maybe one shouldn’t call it bitcoin mining, but the bitcoin lottery ;)
3
Replying to @stadtlegende
Coming back to this, essentially bitcoin is based on hordes of computers throwing dice waiting for a specific outcome.
1
So with #hackweek done, the favorite thing somebody said to me was „I didn’t know you could code.“ #beingalead
10
Wait what, the mining in bitcoins is brute force looking for a key whose hash is smaller than some arbitrary threshold?
6
@SpectralFilter And then the plethora of libraries. I specifically thought I didn‘t like the language, but this time, everything was smooth... ;)
I think this week was officially the most fun I had with Python. And I used to think I don‘t particularly like the language ;) #hackweek
5
Replying to @DJCordhose
Yes, that, and talking to customers. Although I'd say this is not "ML in companies" but "ML in the real world" ;)
2
In research it was always „I have an algorithm, do you know of any good data sets?“ In Zalando it‘s the other way round ;) #hackweek
2 20
Replying to @munterluggauer
One week per year Zalando does a Hackweek where you can work on your favorite project.
Replying to @baggerspion
@therealpadams Still strying to scan those QR codes - that lead to websites?
2
Replying to @baggerspion
@therealpadams Oh so they finally did decide to do something! Well done!
RT @fchollet: Three challenges in our field Hype -- Don't oversell. Talk about what you know. Ethics -- Be aware of the potential for evil…
Replying to @JuanpaMF
@zacharylipton @ryotat It's all useful, just saying depending on how much you can "encode" formally and use in a mathematical proof. So in a PAC setting, can you describe what makes the prob distr "nice" and then use that to prove something?
RT @suzatweet: --- An Adversarial Review of “Adversarial Generation of Natural Language” Or, for fucks sake, DL people, leave language alon…
Replying to @DRMacIver
Yeah. Don't know what the context here is, but I'd also say in most of the cases, everybody is trying to do the right thing locally, but still... .
2
Replying to @rwhitcomb
I guess somewhere the next (final?) stage is to have people buy that stuff for you, btw.
1
Replying to @rwhitcomb
Yeah, the willpower it takes to just buy what you planned to. These are the stories you share with your grandchildren... ;)
1
Replying to @rwhitcomb
Maybe some of the appeal is also that these are usually young people that are still figuring life out.
Replying to @rwhitcomb
God, the assembly. Yeah, I guess at some point you move on to pricier offerings...
Replying to @rwhitcomb
How does West Elm compare to Ikea? We don‘t have those in Europe...
Replying to @rkeytacked
Now that you mention it, the couple I saw that was almost fighting actually was close to the checkout ;)
Replying to @j_c_cabrejas
No, I was there alone. But boy did I see a plethora of human behavior patterns. ;)
Replying to @alansaid
What makes it feel standardized test like is that it practically enforces that everyone takes the same journey through the store.
RT @kade_walker: High school: College is no joke, your instructors are very serious and you have to always be proffesional and respectful.…
IKEA is the place to go to study humans in literally all stages of their relationships.
3 31
Replying to @earino
I‘ve always assumed getting the projector to work for your thesis defense is the real test.
1 1
Replying to @zacharylipton
A lot of the theory we have are based on worst case assumptions about the data, or methods simple enough to know how the result depends on the data. In other words: it‘s hard. #mytwocents
3
Replying to @zacharylipton
@ryotat My idea about this is that you need to have static properties of data to be able to prove something, but part of the cool thing about ML is that we don‘t know how to formally describe the data - and yet it works ;)
3
Replying to @zacharylipton
@ryotat I remember to once have had a hilarious conversation with him at the Gatsby party - but I‘ve always had the hunch that math has issues to produce with data related results.
2
Replying to @Plinz
@zerology Back in the days I thought someone accidentally computing the Ackermann function on a global computing infrastructure would lead to this, but well...
1
RT @Nintendo_Legend: Retweet if you've ever met someone face-to-face after previously only knowing them from Twitter.
RT @hardmaru: You know AI is a bubble when Intel invites Flo Rida to perform at their NIPS party https://t.co/Psyqlt8vcl
Media
Btw, me and @Danzie314 have always been joking that the true „A.I.“ is „K.I.“ - künstliche Intelligenz - the unspoken technology Hinton dreams of at night.
Replying to @agibsonccc
Yeah also not that sure about the „every dev“ part. The technology is fundamentally different than your usual coding. And then there‘s the whole „data driven“ thing. It is a completely different workflow.
Replying to @purbon
To be fair, at least it is not some enforced lock-in, but once you start using aws specific services (and I think you should), you can‘t just migrate away.
1
Wow, AWS just announced video analysis, and a full speech to text, translate and NLP suite.
1 4
Replying to @baggerspion
@therealpadams Yeah, get there by 7am, queue in line to secure a seat.
Two things: 1. when I started with ML back in the 90s I would have never thought I‘d become a commodity like this. 2. just as the keynote comes to this topic, the live feed breaks down.
Media
4 4 View thread →
Replying to @baggerspion
@therealpadams I could just keep myself from doing so. When I arrived they were already rerouting the masses.
So I‘m the second overflow area to watch the reinvent keynote. Starting to feel like a t2.micro instance on an overcrowded server.
1 8
I gotta say, the scale of #reinvent2017 is insane. It sorta works out and you get to see what you registered for, but between venue transfers, waiting in walk-up lines, and just Vegas, I‘m feeling a bit exhausted.
4
Logistics of this event are insane. Also, colleague just registered for session, got a seat while I registered weeks ago and am waiting for walk-up. #eventualconsistency #reinvent2017
View thread →
Replying to @reInventParties
It is! Already spent the morning looking for a place to have breakfast. The guide of guides page is awesome, too! ;)
RT @ebowman: Closing down Black Friday 2017 with @bocytko & @ThorbyG /cc @ZalandoTech https://t.co/iLnVuEe3CT
Media
RT @therealpadams: Remember that single malt from earlier? #ecommerce #blackfriday #awstamtookthephoto https://t.co/yIO0WOMVK1
Media
RT @therealpadams: To everyone working in #ecommerce and especially my sisters and brothers at @ZalandoTech, I wish you a good night’s slee…
Replying to @martinkl
I was wondering that they call people at all. I guess email too easy to ignore ;)
„I‘m really bad at Smalltalk“ - „Yeah I also prefer something with more Clojure.“
1 9
Replying to @agibsonccc
Regarding VCs, I think their goal is make a good exit, totally legit, but the most probably road for that is being acquired by Google et al., who are (I assume) already far ahead in terms of "AI integration".
1
Replying to @agibsonccc
Yes, everything the hype kicked off, "AI" the tech and "AI" the cool shiny new thing have become indistinguishable. I wish more people were aware how "alien" ML's world view is to what is "normal" CS.
2
Replying to @agibsonccc
Big companies necessarily split the work & architecture between many teams. Data is usually not the first thing you think about, but features.
1
Latent Semantic Hashing, Locality Discriminating Allocation, just too many similar sounding acronmys in ML... .
1 4
RT @justintarte: The top two things most people complain about: 1. The way things are. 2. Change.
Winter is when you're still at work and Night Shift already started to change the display's color temperature.
1 6
Replying to @fs111
I was about to fav this but then I saw that I have a circle where there used to be the character count and that made me wonder but what other way to try out than to try to write something long but years of Twitter have tought myself to keep it short #notworthit
View thread →
Replying to @mesirii
@thinkberg Ja, streamdrill wurde zugemacht... . "Best ever" hört sich sehr fundiert an ;), das Paper sieht aber interessant aus!
Replying to @andrew_clegg
@agibsonccc Hehe. When it comes to Japan, nothing surprises me. Do you know they haf ubiquitous email on phones before smartphones?
1
@xaprb Yeah that's because these machines don't "see" the world how we do them, and we only assume that they do.
RT @airberlin: 27.10.2017, arrival of AB6210 #MUC-#TXL. GOODBYE all fans and guests! 😢🖖♥️#HistoryOfairberlin #Forever39 #airberlin 📸by Günt…
RT @therealpadams: They pay us for this. We are hiring! #OSSummit @FokusMan @ZalandoTech https://t.co/wx1IuD1w0l
Replying to @baggerspion
@therealpadams @crateio @Azure @ZalandoTech @claus__m Love the T-Shirt(s)! ;)
Replying to @ChengSoonOng
@teoliphant Still remembering fondly when we had you at our workshop at NIPS. Seems like ages ago!
3
Replying to @shuheikagawa
Beware! What they say: if you learn a skill while drunk, you need to be drunk always to recall!! ;)
1
Replying to @pavlobaron
.. maybe they think the machine gets all that additional connotation attached to seeing another human being - which it doesn't!
Replying to @pavlobaron
Exactly, but humans have this tendency to attribute more "intelligence" to what machines are doing. So when there's a box saying "person"..
Replying to @pavlobaron
I'm suspecting some kind of "visual ELIZA effect". Every time they see a YOLO demo, people probably think that the machien really "gets" it
Why does it seem easier to talk about artificial intelligence when we don‘t even know precisely what intelligence is?
3 9
RT @tanyacash21: @listentojohan @mikiobraun @drewconway might've been marked private, here it is! https://t.co/Bxrw73EPdu
RT @bigdata: Daniel Kang @stanford describes NoScope a system for fast querying of video #StrataData #hardcoredatascience https://t.co/tnA2…
Interesting overview by Sarah Manning on how Etsy is driving a culture of experimentation. #StrataData
Replying to @BBende
@suneelmarthi Same here, iphone claimed it lost battery percents by the minute. Then continued to work at 1% for 4h 🤷🏻‍♂️
Replying to @stefan_will
I have found recommendations how to recalibrate, but those have been somewhat... esoteric ;)
Replying to @stefan_will
Yeah, I hope calibration. Already installed the GM on Sunday, reindexing should be through I guess.
So first I thought after the ios11 update my phone’s battery life was pretty bad. Now my phone has been at 1% for the past 4h... 📱🤔⁉️
Replying to @Quesada
@gergrubi I just checked, what a graveyard of startups... foursquare, vine, ..., TWIMPACT, ... ;)
RT @soumithchintala: NIPS Conference Registrations 2002 thru 2019. [2018] War erupts for tickets [2019] AI researchers discover time travel…
Replying to @baggerspion
@therealpadams @ZalandoTech (1) as a lead, taking the blame is part of my job description, so no problem here.
1
The worst part about design by committee is that it is all critics and too few creators.
2 9
Replying to @Quesada
@gergrubi I think you can go into Settings > Apps and revoke access.
Replying to @fs111
Oooh yeah. Maybe chat messages should autoexpite after ten minutes, then people would understand this better.
Replying to @emschreuder
To be honest, nobody asked me, or it got lost in my email flood ;)
RT @antoine77340: The hype of being an AI research scientist nowadays. https://t.co/karJtj6yIG
Media
Replying to @oldJavaGuy
"How do I mute this?" - "Here is how you mute this" - "Stop replying to all!!"
RT @vambenepe: Today at work someone mistyped "Google" as "Goodle" and for the rest of the meeting we talked about the "Good Old Cloud Plat…
RT @DataScienceLA: If you do #kaggle to learn #machinelearning, you are missing on 80% of things you need for ML in real life/production ht…
RT @ScottGustin: International Space Station #Eclipse photobomb via @NASA https://t.co/NashX5OCEX
Media Media
Replying to @fs111
I meant "I miss those days, too" Would edit. If I could. At least that stayed the same ;)
As North America sleeps my timeline dreams of better days when it was just about the next enterprise bandwagon ;)
4
Replying to @shuheikagawa
@unsoundscapes I spent countless afternoons in my childhood designing monochrome 8x8 sprites on Commodore home computers.
3
Replying to @arstechnica
@azeem @DrPizza Still waiting for the "search tab titles" feature.
Me: "Here, have some cookies" Kid pulls out phone: "Let me share with my friends!" Me: "Kid, don't let them hijack your session" #dadjoke #😅
4
Replying to @philgyford
@GennHutchison Also just recently finish Last of Us. That was epic!
@LauritaApplez @therealpadams @birgitta410 @jpetazzo @try_except_ @valgog @MaineC @janl I think I already know everyone except for @birgitta410 and @jpetazzo!
Forget the analogy gap. If we get an AI that can detect as accurately as my kids whether I really mean it, then we've made process.
1 3
@LauritaApplez "Diet Dr. Pepper manages to combine 23 distinct flavors in a way that destroys every single one."
Replying to @egrefen
Now thinking of it, I have "shut down" numerous "AIs" who I thought hadn't performed well, but actually I stopped the AI uprising!
3
RT @robertoglezcano: If you cannot explain something in simple terms, you don't understand it. —Prof. Richard Feynman https://t.co/vBcNIVmr…
Two years with @ZalandoTech today! 🎂🙌 First act: check which internal access roles need renewal ;)
14
RT @benhamner: Replace "AI" with "matrix multiplication & gradient descent" in the calls for "government regulation of AI" to see just how…
Replying to @SwiftOnSecurity
I was honestly looking for seasonal effects of fedora hat buying cycles...
RT @slavivanov: What do you do if your #ML model is not working? Here are 37 ways to debug an unruly neural network (by yours truly) https:…
RT @therealpadams: Day of riverboating with the almighty Teams Poirot and Boost of @ZalandoTech. https://t.co/7dmA5K4fRP
Media
RT @saywhatagain: Ahhh, the Law of Millennial Trendpiece Conservation at work https://t.co/dDEQKYw9fj
Media Media
Replying to @kirel
What I meant was that I'd already be happy if we managed to get a dog without all the speech, planning, and self awareness on top.
1
Replying to @kirel
I would say I don't know enough about how that stuff actially works to be able to say we build it once and it will scale.
1
Replying to @kirel
Apparently there are huge gaps in observable intelligence, but yeah it seems to be made of the same stuff.
Replying to @kirel
Actually I think it is even harder than what we're working on right now, but still easier than human level AI.
Instead of going for human level AI, we should probably start with dog level AI. And cuteness 🐶
1 18
RT @CodeWisdom: "First you learn the value of abstraction, then you learn the cost of abstraction, then you're ready to engineer." - Kent B…
Replying to @superglaze
Won an iphone already today? I really don't understand why websites let that do with themselves.
1
Replying to @munterluggauer
Hm. Depends a lot on the algorithm and so on, but it is certainly doable ;)
A math TA once told me she always starts with a blank sheet of paper and writes down "what do we know" and "what do we want".
8
RT @stilkov: This running commentary of watching "Alien" for the first time is the most hilarious thing I've read in a while https://t.co/y…
Replying to @jorgejbla
@bbchase Oh great, got an ad covering the whole page which looks like it comes from Google telling me I won the lottery on that article.
1
RT @ssdpd: Another cloud comes and asks the first “may I join, mate”. That's how club mate came to the internet startups. https://t.co/UsqZ…
RT @CodeWisdom: "We build our computer (systems) the way we build our cities: over time, without a plan, on top of ruins." - Ellen Ullman
RT @andrew_clegg: Quote in article: "there was an instance where black swans were interfering" - does she mean that literally or is it a st…
Alright now I think it has been raining for 24h straight here in Berlin... 🌧☔️🙄🌧
3
Replying to @ABridgwater
@nvidia From the logo I would even say it should be spelled nVIDIA ;-P
1
Replying to @martinkl
I agree. I went through lengths to buy from them directly for exactly that reason... .
1
Replying to @hanchenxiong
It is a mix of just mock services and reimplementation of some core APIs (for example for kinesis and dynamoDB)
Replying to @truemped
@mhausenblas Yeah, and does its Route53 support cross-region failover?
1
@LauritaApplez @otrosien "Do people still test locally" becomes the B side of "Nobody thinks about the door"
Replying to @fs111
Deep Learning for fraud detection on Blockchain transaction Kafka stream, got it!
Replying to @mhausenblas
I think proprietary cloud services are hard to replicate locally but everything else should be possible to deploy with docker etc., right?
1
@klabol Guys! Can't you see where this is heading?!? #skynet Also: what about #nofreelunch?
RT @chasejarvis: Are you running TO something or AWAY from something? The difference in mindset is a game changer. 🔑🔑🔑 https://t.co/Vrgnvdw…
Replying to @truemped
Deep Learning is the one exception, of course. Especially if you use tensorflow. ;)
1
Oddly, the new Twitter UI manages to elicit the feeling something is happening, something is new, and makes me get back to Twitter more.
5
RT @seanjtaylor: In my experience, regularizing linear models is clearly a case of the ridge get richer.
RT @beaucronin: Not to pick a fight with an anonymous conference attendee on the other side of the planet, but: yes, you can https://t.co/M…
And again... #bbuzz Talking with @alaa_elhadba about the modern architecture of search.
Media
2 8
@LauritaApplez Btw, checked it myself and now I see what you mean. Twix are there but machine claims there are non. Probably won't fix itself.
Found this in jblas's commit log. Best commit message ever... 😂 #getmeout #howdoiquitthis
Media
2 10
@LauritaApplez @kode4food No way to get a head start on technical debt like a lengthy document rich onboarding process 😜
Replying to @shuheikagawa
Maybe I need to tweak the routing such that my old disqus plugins still work, though o_O
Replying to @shuheikagawa
Hm don't know, to be honest, I just ran the 'yarn run init' as said in the docs ;)
Replying to @agibsonccc
@suneelmarthi @deeplearning4j And then you can write a nice wrapper if you wish to. Fully agree! ;)
1
Replying to @fs111
To be honest, I'm kinda wishing for those nice markdown wysiwyg editors like medium... 😅
1
RT @Reza_Zadeh: NIPS deadline was May 19. AWS price for p2.16xlarge (each with 16 GPUs) spiked to its maximum possible ($144/hr) two days p…
When customer care says "and be sure to clear your browser caches" and you're like "let me tell you about 502 from nginx"
2 6
Replying to @roidrage
@LauritaApplez @thedailybeast TBH this discussion is already veeeery suspicous! ;)
RT @ilnumerics: For our german fellows: @mikiobraun lifting secrets about #deeplearning and #bigdata buzz #parallel2017 #ilnumerics https:/…
RT @JoshuaDroid: This table has 118 elements and number 47 will SHOCK you! ..because silver has the highest electrical conductivity. shoc…
New idea: Coding for Managers, when you only have one hour and don't want to spend it all on setting up the project config.
1 4
@LauritaApplez @try_except_ @JanMussler @timkrgr Hello? Everything else would be superrude!
2
Free startup ideas: (1) video conferencing that works, (2) agile planning tool that doesn't suck, (3) mail client that works like Twitter
1 9
RT @aCraigPfeifer: @ChrisDiehl I'm still skeptical about merging my computer with my toaster, let alone my brain.
Replying to @ChrisDiehl
@aCraigPfeifer Do you guys also wonder sometimes wth is going on right now? ;)
1
Replying to @lojikil
proudly born in '75. First computer was a barely working zx-81. Have seen thing you people wouldn't believe ;)
1
@kode4food I've been at a parallel computing and HPC computering conference a couple of weeks ago, and they do, interestingly.
While trying to explain to my kids what deep learning is I realized I worked as a scientist in the area of thinking machines. #prettycooleh
3
RT @robertoglezcano: Actual PhD students are not so jubilant. @sketchscience https://t.co/ju4wug8Pws
Media
RT @danielsing3r: I visited @ThePracticalDev today. Was not disappointed. https://t.co/tW0P0VNYpN
Media
RT @EmmyA2: carry yourself with the confidence of a ̶w̶h̶i̶t̶e̶ ̶m̶a̶n̶ girl in dopeass glasses busting into a white man's meeting https:/…
Hm. According to Google Scholar, the past two years (after I left academia) were the most successful in terms of citations 🤷🏻‍♂️
6
Sooo apparently N. Virginia is dealing with either an alien invasion or a zombie outbreak.
1
Maybe I should switch to a pay per meeting kind of compensation model. Maybe this gives the wrong incentive ;)
3
Just because you are using the JVM as an execution backend for your product doesn't mean stacktraces make good error messages.
2
RT @wesgeorge: The best thing about UDP jokes is that I don't care if you get it. https://t.co/hKJfGcJuug
Media
RT @DRMacIver: I guess it's easy to say this in retrospect, but it's not *very* surprising that computers are better than humans at structu…
TFW you have to switch from Week to Day view in the calendar to be able to read all those double booked meetings. I meant triple booked.
3
RT @ivan0yu: One morning, as Franz Kafka was waking up from anxious dreams, he discovered that in bed he had been changed into a distribute…
Replying to @DRMacIver
oh I mean 1/time = power? Sorry, my physics training is decades old by now ;)
Replying to @fs111
oh yeah. Their solution: use greek scripts/bold/overlined/underlined/overarowed symbols if you run out of letters.
Replying to @fs111
ha, hadoop... stone age... there's a joke waiting to be told somewhere in there ;)
Replying to @semithinking
yeah! I really like how todos are embedded in the overall structure. Anything else is too flat for me.
Replying to @hesk
@truemped I guess you could in principle run some texteditor on iCloud or Dropbox files for data entry... ;)
Replying to @truemped
@hesk there is an app but it is super old. Like pre-Retina. On mobile!
Replying to @Springcoil
haven't tried that. But org mode's combination of a foldable hierarchical bullet list plus todos and calendar... ;)
Replying to @hesk
unfortunately there isn't one. Still calling it an "app", because that's what we do right now, right?
So turns out non-tech people find my Cloudera "Data is the new bacon" T-shirt hillarious (too).
4
Replying to @truemped
@ivan0yu if you write lisp, the editor needs to be written in lisp, too ;)
@moellus ja, aber noch scheue ich das Licht der Öffentlichkeit #hastenixbessereszutunfragen
Curse you, @truemped to introduce me to clojure. I feel like I can tell nobody the extent of professional thought share, but it's late.
3
RT @haldaume3: No offense, but I gotta poke fun here. The fact that in 2 yrs, AI has been reduced to classification challenges is sadly hil…