Fatma

one of the most important things I know about deep learning I learned from this paper: "Pretraining Without Attention" this what I found so surprising: these people developed an architecture very different from Transformers called BiGS, spent months and months optimizing it and training different configurations, only to discover that at the same parameter count, a wildly different architecture produces identical performance to transformers this may imply that as long as there are enough parameters, and things are reasonably well-conditioned (i.e. a decent number of nonlinearities and and connections between the pieces) then it really doesn't matter how you arrange them, i.e. any sufficiently good architecture works just fine i feel there's something really deep here, and we may be already very close to the upper bound of how well we can approximate a given function given a certain amount of compute. so we should spend more time thinking about other questions, such as what that function should actually look like (what data? which objective function?) and how to make it more efficient

Silence on Gaza is no longer an option - after seeing those images, really seeing them, you still want to stay “neutral”, stay silent, or talk about “Islamist oppression” and “LGBT intolerance”, don’t call yourself a feminist. Because you are not one.

Here's a collection of my favorite #ICCV2023 papers on diffusion-based image editing. (1/4)

Creating my personal memory aid thread to bookmark the standout papers. Can't wait to dive thoroughly into these later! 🤓

My lecture notes on regularization techniques in machine learning dropbox.com/scl/fi/ivyb3ws… Check out the section on double descent suggested by @ylecun :)

Sigmoid Loss for Language Image Pre-Training paper (CLIP with sigmoid loss instead of softmax ) x.com/giffmana/statu…

CroCo v2: Improved Cross-view Completion Pre-training for Stereo Matching and Optical Flow x.com/chriswolfvisio…

vision transformers need registers x.com/TimDarcet/stat…

Llamas are here

Let’s go 🤗🤗🤗#ICCV2023

@TiagodeVassal @ylecun @aerinykim I’m not talking about selfies, just wanted to highlight the percentage of women attendees . While it’s still low , it's promising to see that it's on an upward trajectory compared to previous years

Selfied after talking with students in front of the Meta booth at ICCV - Paris

@ylecun @aerinykim 18% of participants were women . (The statistics were shared during the conference )

@aerinykim I'd say there were at least as many women as men asking for selfies. But in this crowded picture, you only see tall men.

Cartoon-like filter at the TikTok booth @ #ICCV2023

I've been told the traffic in Paris will be horrible, I didn't realize they were referring to the #ICCV poster session.

#CVPR2023 is mostly posters this year, which is good! Poster sessions can be great. But a lot of people are not so great at presenting their poster. Here's some of the advice I give grad students when preparing for poster sessions.

“Make yourself be lucky!”- Alyosha Efros