Clementine Domine 🍊 @CCN
@clementinedomine.bsky.social
•PhD student @ https://www.ucl.ac.uk/gatsby 🧠💻
•Masters Theoretical Physics UoM|UCLA🪐
•Intern @zuckermanbrain.bsky.social|
@SapienzaRoma | @CERN | @EPFL
https://linktr.ee/Clementine_Domine
•Masters Theoretical Physics UoM|UCLA🪐
•Intern @zuckermanbrain.bsky.social|
@SapienzaRoma | @CERN | @EPFL
https://linktr.ee/Clementine_Domine
Reposted by Clementine Domine 🍊 @CCN
👉 Join our Google group for more updates: groups.google.com/u/2/g/ellis-...
ELLIS UniReps Speaker Series - Google Groups
groups.google.com
May 20, 2025 at 11:57 AM
👉 Join our Google group for more updates: groups.google.com/u/2/g/ellis-...
Thank you to my amazing co-authors: Yoonsoo Nam, Seok Hyeong Lee, Yeachan Park, Charles London, Wonyl Choi, Niclas Alexander Göring, Seungjai Lee 4/4
May 5, 2025 at 4:37 PM
Thank you to my amazing co-authors: Yoonsoo Nam, Seok Hyeong Lee, Yeachan Park, Charles London, Wonyl Choi, Niclas Alexander Göring, Seungjai Lee 4/4
💬 So next time a reviewer says, “Linear networks aren’t used in practice,” you’ll know where to point them. 3/4
May 5, 2025 at 4:37 PM
💬 So next time a reviewer says, “Linear networks aren’t used in practice,” you’ll know where to point them. 3/4
Surprisingly, layerwise linear models (no nonlinearities!) can already explain key deep learning behaviors:
– Neural Collapse
– Emergence
– Lazy vs. Rich Regimes
– Grokking
2/4
– Neural Collapse
– Emergence
– Lazy vs. Rich Regimes
– Grokking
2/4
May 5, 2025 at 4:37 PM
Surprisingly, layerwise linear models (no nonlinearities!) can already explain key deep learning behaviors:
– Neural Collapse
– Emergence
– Lazy vs. Rich Regimes
– Grokking
2/4
– Neural Collapse
– Emergence
– Lazy vs. Rich Regimes
– Grokking
2/4