Thomas Fel
@Napoolar
Followers
1K
Following
9K
Media
60
Statuses
1K
Explainability, Computer Vision, Neuro-AI. Research Fellow @KempnerInst, @Harvard. Prev. @tserre lab, @Google, @GoPro. Crêpe lover.
Boston, MA
Joined February 2017
Train your vision SAE on Monday, then again on Tuesday, and you'll find only about 30% of the learned concepts match. ⚓ We propose Archetypal SAE which anchors concepts in the real data’s convex hull, delivering stable and consistent dictionaries.
6
79
352
RT @KempnerInst: Interested in the latest work from the #KempnerInstitute? Check out papers and preprints from June's Research Roundup. h….
0
3
0
Check out this amazing work by @fenildoshi009 on shape holistic processing in vision models! 🍀.
🧵 What if two images have the same local parts but represent different global shapes purely through part arrangement? Humans can spot the difference instantly! The question is can vision models do the same? . 1/15
1
4
34
RT @fenildoshi009: 💡 A striking U-shaped pattern emerged. In DINOv2, ablating long-range attention had minimal effect on early & late layer….
0
1
0
RT @fenildoshi009: To test this, we surgically ablate attention in ViTs. We restrict each patch to:. 🟦 “Attend Inside” (block long-range in….
0
1
0
RT @EkdeepL: This collab was one of the most beautiful papers I've ever worked on! The amount I learned from @danielwurgaft was insane and….
0
2
0
RT @Cohere_Labs: Scaling Self-Supervised Learning for Vision🧠🖼️. @TimDarcet will present on July 6th and he'll cover contrastive learning,….
0
2
0
RT @chingfang17: Humans and animals can rapidly learn in new environments. What computations support this? We study the mechanisms of in-co….
0
55
0
RT @KempnerInst: The Kempner Institute congratulates its research fellows Isabel Papadimitriou (@isabelpapad) and Jenn Hu (@_jennhu) for th….
0
5
0
RT @isabelpapad: Check out our ACL paper! We use shapley interactions to see which words (and phones) interact non-linearly -- what we lose….
0
7
0
RT @ema_marconato: 🧵Why are linear properties so ubiquitous in LLM representations?. We explore this question through the lens of 𝗶𝗱𝗲𝗻𝘁𝗶𝗳𝗶𝗮….
0
61
0
RT @KempnerInst: NEW: George Alvarez (@grez72) of @PsychHarvard and the #KempnerInstitute shows how long-range feedback projections can enh….
0
2
0
RT @s_scardapane: *Universal Sparse Autoencoders*.by @HThasarathan @Napoolar @MatthewKowal9 @CSProfKGD . They train a shared SAE latent spa….
0
38
0
RT @KempnerInst: NEW: Asma Ghandeharioun (@ghandeharioun) of @GoogleDeepMind demonstrates techniques to enhance interpretability of large l….
0
3
0
Around CVPR for the next 2 days—if you're into interpretability, SAEs, complexity, or just wanna know how cool @KempnerInst is, hit me up 👋
0
9
70
RT @SuryaGanguli: Looking forward to speaking today at Harvard’s @KempnerInst NeuroAI conference on theories of learning, creativity and re….
0
19
0
RT @raphaelsrty: I'm thrilled to announce the release of FastPlaid ! 🚀🚀. FastPlaid is a high-performance engine for multi-vector search, bu….
0
36
0