🔥Telegram İfşa

We are glad to announce the talk series on AI for sustainability (AI4S) organized by the ETH AI Center. Check out our website ai.ethz.ch/research/talk-….

I am happy to announce our new preprint arxiv.org/pdf/2203.03597…! We discuss why it is crucial for interpolating models to carefully choose a moderate inductive bias and show that this can even lead to close to minimax optimal rates.

How does the ability of interpolators to generalize on noisy data depend on the strength of their inductive biases? We are delighted to present our recent work on this question today at the ICML Poster #1109. See also proceedings.mlr.press/v162/donhauser…

We are presenting at 5 posters in #ICML2022 workshops with amazing collaborators. 3 of them today (22nd) and 2 tomorrow (23rd). See below for details 👇

Uniform versus uncertainty sampling: When being active is less efficient than staying passive shorturl.at/dmOY4 - excited to hear your thoughts on our theoretical and empirical findings for small budget and high-dimensional #activelearning at the #ReALML workshop #icml2022.

(1/5) LLMs risk memorizing and regurgitating training data, raising copyright concerns. Our new work introduces CP-Fuse, a strategy to fuse LLMs trained on disjoint sets of protected material. The goal? Preventing unintended regurgitation 🧵 Paper: arxiv.org/pdf/2412.06619

Using mechanistic #interpretability 💻 to advance scientific #discovery 🧪 & capture striking biology? 🧬 Come see Jason Hartford's oral presentation 👨‍🏫 @ #NeurIPS2024 Interpretable AI workshop 🦾 to learn about extracting features from large 🔬 MAEs! 📄: openreview.net/forum?id=jYl7k…

I am particularly bullish on using mechanistic interpretability (especially SAEs) to better understand and discover new knowledge about biology and medicine. "Towards scientific discovery with dictionary learning: Extracting biological concepts from microscopy foundation models"