Worry about Natural Intelligence now and AIs won't be a problem later.
Computer Vision, mostly, but multi-curious.
Opinions, comments, etc. are my own.
Do Vision Transformers See Like Convolutional Neural Networks?
New paper https://t.co/mxLCIRBRLy
The successes of Transformers in computer vision prompts a fundamental question: how are they solving these tasks? Do Transformers act like CNNs, or learn very different features?
@_KarenHao@strwbilly "Many of the problems that were uncovered are linked to the poor quality of the data that researchers used to develop their tools."
If big amounts of quality data is not freely available to all researchers this is what you get. No data no models.
"Thereโs not been enough focus on the real problem, which is building planetary-scale machine learningโbased systems that actually work, deliver value to humans, and do not amplify inequities.โ๐ and ๐
Mike Jordan defending ML engineering! Good engineering gave us the GPU convnets, the transformers, torch, numpy, etc. The popular diminishing meme โitโs-just-engineeringโ is silly, and holds us back. I โค๏ธ creative, rigorous, robust, safe engineering. https://t.co/4GcTjf0d0F
Help me out in a Silly Sunday Squabble ๐! Say you want to fake a model file, a dummy: 1) as small as possible on disk, 2) torch.load(dummy) must not crash. (It's ok predictions are meaningless.) What is the smallest such file? Can one avoid pickling the weights in a dummy?
Chuckling at the irony of having that "Thirfty Machine Learning" talk open beside a tab with a paper on "Large Batch Optimization For Deep Learning: Training Bert In 76 Minutes": https://t.co/nJfDgsHb6X
๐
#dontjudgeme
I could pin every slide on the (virtual) office walls. @RebeccaBilbro delivering common sense in screenshot-able doses in "Thrifty Machine Learning" at #PyDataGlobal2020.
https://t.co/OQ4rWMRh9T
I'm seriously loving @gather_town. The feel to it - and some usability details. That tiny preview popping up as you walk beside a poster ๐
#pydataGlobal2020
Today #PyDataGlobal2020 kicked off!
Don't online conferences have a bigger FOMO factor?
In person, you make a choice and accept (mostly) you won't see everything.
Online though...?
A related problem: the huge lack of diversity in the field of artificial intelligence. Even an academic project with art as its main application matters & should not have gone all the way to publication before someone noticed that it was hugely biased. https://t.co/wN6UCLlSRd
An image of @BarackObama getting upsampled into a white guy is floating around because it illustrates racial bias in #MachineLearning. Just in case you think it isn't real, it is, I got the code working locally. Here is me, and here is @AOC.
Now that the #CVPR2020 site is back up and running, please tune in for our first sessions! Remember that authors will be participating in Live Q&As during the session. Links in thread.
Sooo, while we're waiting for the #cvpr2020 site to recover, here's one of the open Tutorials ๐
"Visual Recognition for Images, Video, and 3D"
https://t.co/YbhguEmZQs
"will take place unless safety/health regulations requires that it be cancelled"
"current large event ban in Seattle (...) will likely be extended on a rolling basis"
Comments' hint: they likely can't spell out they'll get cancelled, don't book your ticket/hotel just yet.
Silver lining? Companies could come out of the #coronavirus quarantines with better remote collaboration practices and tools. Employees (managers included) could grow into a better remote work culture. #stayathome
Chocolate is good, but revisiting your TF 1.7 code from 2 years ago, upgrading to 1.15, auto-replacing import strings and seeing it train correctly + all tests green is... another level.
The step to 2.0 won't be this easy, so I'll just savour this a few minutes.