I want to understand life. This new project is my biggest advance yet 🚨
Announcing: SE(3) Equivariant Transformer Cellular Automata for Morphogenesis
00:00
Cameron Witkowski
1,398 posts
Cameron Witkowski
@witkowski_cam
co-founder openlens.com
- Replying to @witkowski_camCNNs (@ylecun, @Yoshua_Bengio) led to Spherical CNNs (@TacoCohen, @mario1geiger, @jonkhler, @wellingmax), and eventually to the SE(3) Transformer (@FabianFuchsML, @danielewworrall, Volker Fischer, @wellingmax). SE(3) is the special euclidean group, consisting of rigid
- Replying to @langofmindAre you telling me the year 0001 is equally likely as 2024?
- Replying to @witkowski_camMoving from 2D to 3D, and fixed grids to free-moving cells required the insights of geometric deep learning, spearheaded by @mmbronstein, @joanbruna, @TacoCohen, and @PetarV_93 and beautifully outlined in this proto-book: arxiv.org/abs/2104.13478. Geometric deep learning
- Replying to @eshearDid you know technology, text, and fabric (textiles) all share the same root?
- Had not heard of context distillation when we wrote the paper back in 2024 but this is great stuff & way ahead of its time! Our initial paper showed us that prompts could in principle be converted into weight updates — and surprisingly fast with new advances like LoRA, andJust read their paper. Looks like they re-invented an existing method known as context distillation (or merely re-branded it for their startup). No mention of prior work, sadly. Links to papers in thread.
- Replying to @witkowski_camA big inspiration for this project was the paper “Growing Neural Cellular Automata” by @zzznah, @RandazzoEttore, @eyvindn, @drmichaellevin. They put a twist on Von Neumann cellular automata by demonstrating the interaction rules could be learned using neural networks to
- Language Models can simulate intelligence. How is simulated intelligence different from actual intelligence? Well, how is fiction different from truth? From an amazing conversation with @ecsquendor and @ABhargava2000 on the @MLStreetTalk podcast! 🧵(1/10)
00:00 - Replying to @witkowski_camThe video of the frog developing was taken by Jan Van Ijken (janvanijken.com). Other inspirations and references include: @hardmaru, @ElowitzLab, @geoffreyhinton, Karl Friston, John Conway, Carver Mead, Richard Feynman, Eric Winfree, John Hopfield. I briefly foreshadowed
- Why does erasing information necessarily cost energy, regardless of how it is represented? I answer this in my recent paper published in Entropy: mdpi.com/1099-4300/26/3… The story begins with Maxwell's Demon... 1/n
- Sneak peek of my poster for Quantum Thermodynamics Conference 2024 at @UofMaryland next week! The central circle contains the critical intuition—if you understand that, everything else falls into place! ... including an original proof of Landauer's principle!
- A short(er) explanation of SE(3) Equivariant Transformer Cellular Automata for Morphogenesis Credits to @ABhargava2000 for the video and for some major recent breakthroughs
00:00 - Replying to @witkowski_camThis video highlights my ongoing MASc Thesis at @UofT in the department of ECE @eceuoft, supervised by Prof. Stephen Brown and Prof. Kevin Truong. Big thank you to @ABhargava2000, a PhD candidate at @Caltech CNS in the Thomson group, for editing this video, inspiring many
- Replying to @skdhIf entanglement is just correlation and measurement really “does nothing,” then Bell violations imply something else has to give—namely, statistical independence. That’s fine—if you’re invoking superdeterminism. But that’s a heavy assumption. Shouldn’t we at least name the cost?
