How Facebook scales AI: a nice piece at ZDNet following a keynote at the International Symposium on Computer Architecture (ISCA) by Kim Hazelwood, head of Facebook's AI Infrastructure group.
— Yann LeCun (@ylecun) June 7, 2018
"Most of Facebook's two... https://t.co/qVEdQm4D8R
1989: Self-driving car with 4 neurons
— hardmaru (@hardmaru) June 7, 2018
2018: Play Atari game with 6 neurons#progress https://t.co/Yu7aO9lV0B
Even with a sample size greater than what is typical in the field (n = 100), the replicability of fMRI studies is still undesirable low 😟https://t.co/LmuYEwUnA4 pic.twitter.com/GKsKHEL0CT
— Indrajeet Patil (@patilindrajeets) June 7, 2018
The lack of scientific thinking affects every subfield of CS. A recent paper about computer security points out what should have been obvious: we've been trying to theorem-prove our way to scientific knowledge, but we need experiments and falsifiablily. https://t.co/GkMU5SIpxy
— Arvind Narayanan (@random_walker) June 7, 2018
This essay is a must read for everyone thinking about AI predictions. https://t.co/lkzAmF8rBc
— Nando de Freitas (@NandoDF) June 7, 2018
As a data science type person, if there was one thing I could say to most companies, it would be this: Store your data properly, goddammit.
— Emma Vitz (@EmmaVitz) June 7, 2018
We can talk about ML and AI all you like, but if the data's not there or the quality's not there, that's all for nothing.
googled for an error message in pytorch, read my own answer from a year ago. full circle! Then dug around some stats. There are 34,300 posts on the PyTorch forums, viewed 7.6 million times. I wrote 1800 of them. So cool :D
— Soumith Chintala (@soumithchintala) June 7, 2018
Different interpretation of the same results: @jacobeisenstein HMM beats LSTM on small data @sleepinyourhat wow, LSTM beats HMM already with 500 sentences. Proof links: https://t.co/ao80M4NHhR https://t.co/wIxdqm5p93
— Leonid Boytsov (@srchvrs) June 6, 2018
Data scientists do three different jobs in varying combinations:
— Brandon Rohrer (@_brohrer_) June 5, 2018
Analysis – turning raw information into knowledge that can be acted on
Modeling – using the data we have to estimate the data we wish we had
Engineering – making everything else work faster, robustly, and at scale
"From AI to ML to AI: On Swirling Nomenclature & Slurried Thought" - a new installment related to "the AI Misinformation Epidemic" - on @approxcorrect https://t.co/4UrIfJubbI
— Zachary Lipton (@zacharylipton) June 5, 2018
"What do you do?"
— karl rohe (@karlrohe) June 5, 2018
"I'm a Statistician."
"Oh wow, I hated that class."
"Sorry. We are bad at teaching."
"I bet you've seen some crazy statistics!"
“uhhhhh.” https://t.co/JztjgO0TKP
Great long read in @theipaper by @robhastings, looking at how Streams, our secure mobile app, is working at @RoyalFreeNHS. Well worth a read to understand the enormous potential for positive impact of digitisation in healthcare! :)https://t.co/aLnvxZOjHF
— Mustafa Suleyman (@mustafasuleymn) June 5, 2018
