Ekdeep Singh
@EkdeepL
Postdoc at CBS-NTT Program on Physics of Intelligence, Harvard University.
🚨New paper! We know models learn distinct in-context learning strategies, but *why*? Why generalize instead of memorize to lower loss? And why is generalization transient? Our work explains this & *predicts Transformer behavior throughout training* without its weights! 🧵 1/
Really nice analysis!
🚨New paper! We know models learn distinct in-context learning strategies, but *why*? Why generalize instead of memorize to lower loss? And why is generalization transient? Our work explains this & *predicts Transformer behavior throughout training* without its weights! 🧵 1/
Excited to share new work @icmlconf by Loek van Rossem exploring the development of computational algorithms in recurrent neural networks. Hear it live tomorrow, Oral 1D, Tues 14 Jul West Exhibition Hall C: icml.cc/virtual/2025/p… Paper: openreview.net/forum?id=3go0l… (1/11)
Check out my boy @dmkrash presenting our “outstanding paper award” winner at the Actionable Interpretability workshop today!
Check out my posters today if you're at ICML! 1) Detecting high-stakes interactions with activation probes — Outstanding paper @ Actionable interp workshop, 10:40-11:40 2) LLMs’ activations linearly encode training-order recency — Best paper runner up @ MemFM workshop, 2:30-3:45
While you still can, snatch this prodigy undergrad for your lab when he applies for PhDs this fall!
Thank you to everyone who swung by our poster presentation!!! So many engaging conversations today. #ICML2025
Submit to our workshop on contextualizing Cogsci approaches for understanding neural networks---"Cognitive interpretability"!
We’re excited to announce the first workshop on CogInterp: Interpreting Cognition in Deep Learning Models @ NeurIPS 2025! 📣 How can we interpret the algorithms and representations underlying complex behavior in deep learning models? 🌐 coginterp.github.io/neurips2025/ 1/
I'll be at ICML beginning this Monday---hit me up if you'd like to chat!
🚨 New preprint! 🚨 Everyone loves causal interp. It’s coherently defined! It makes testable predictions about mechanistic interventions! But what if we had a different objective: predicting model behavior not under mechanistic interventions, but on unseen input data?
Don't forget to tune in tomorrow, July 10th for a session with @EkdeepL on "Rational Analysis of In-Context Learning Elicits a Loss-Complexity Tradeoff" Learn more: cohere.com/events/Cohere-…
How do LLMs learn new tasks from just a few examples? What’s happening inside during in-context learning? 🤔 Join us July 10 for a talk by @EkdeepL on how LLMs adapt like cognitive maps—and how we can predict their behavior without accessing weights.
I'll be giving an online talk at Cohere Labs---join!
How do LLMs learn new tasks from just a few examples? What’s happening inside during in-context learning? 🤔 Join us July 10 for a talk by @EkdeepL on how LLMs adapt like cognitive maps—and how we can predict their behavior without accessing weights.
Excited to announce our recent work on understanding training-time emergence in Transformers! Thread🧵(1/11)
Bayesian models as the ultimate normative theories neural network as the ultimate task-performing models
It turns out that a lot of the most interesting behavior of LLMs can be explained without knowing anything about architecture or learning algorithms. Here we predict the rise (and fall) of in-context learning using hierarchical Bayesian methods.
Can we record and study human chains of thought? The think-aloud method, where participants voice their thoughts as they solve a task, offers a way! In our #CogSci2025 paper co-led with Ben Prystawski, we introduce a method to automate analysis of human reasoning traces! (1/8)🧵
It turns out that a lot of the most interesting behavior of LLMs can be explained without knowing anything about architecture or learning algorithms. Here we predict the rise (and fall) of in-context learning using hierarchical Bayesian methods.
🚨New paper! We know models learn distinct in-context learning strategies, but *why*? Why generalize instead of memorize to lower loss? And why is generalization transient? Our work explains this & *predicts Transformer behavior throughout training* without its weights! 🧵 1/
🚨 Registration is live! 🚨 The New England Mechanistic Interpretability (NEMI) Workshop is happening August 22nd 2025 at Northeastern University! A chance for the mech interp community to nerd out on how models really work 🧠🤖 🌐 Info: nemiconf.github.io/summer25/ 📝 Register:…
I've struggled to announce this amidst so much dark & awful going on in the world, but with 1mo to go, I wanted to share that: (i) I finally graduated; (ii) In August, I'll begin as an assistant professor in the CS dept. of the National University of Singapore.
Humans and animals can rapidly learn in new environments. What computations support this? We study the mechanisms of in-context reinforcement learning in transformers, and propose how episodic memory can support rapid learning. Work w/ @KanakaRajanPhD: arxiv.org/abs/2506.19686
This collab was one of the most beautiful papers I've ever worked on! The amount I learned from @danielwurgaft was insane and you should follow him to inherit some gems too :D
🚨New paper! We know models learn distinct in-context learning strategies, but *why*? Why generalize instead of memorize to lower loss? And why is generalization transient? Our work explains this & *predicts Transformer behavior throughout training* without its weights! 🧵 1/