🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

1) Can we characterize learnability and design minimax optimal PAC learners for realizable regression? 2) Can we characterize the optimal cumulative loss and design optimal online learners for realizable regression? arxiv.org/abs/2307.03848

Delighted to see several exciting results and colleagues covered in this massive article on Complexity Theory and Meta-Complexity written by Ben Brubaker for Quanta Magazine: quantamagazine.org/complexity-the…

⭐️New Paper⭐️ Can we use game theory to improve language model (LM) decoding? We introduce Equilibrium-Ranking (ER), where we cast LM decoding as an imperfect-information game 👾. On multiple tasks, LLaMA-7B with ER can outperform the much larger LLaMA-65B model. 🧵⬇️1/

Looking forward to talking about this new paper with Ankur Moitra and Dhruv Rohatgi, arxiv.org/abs/2309.09457, in which we give a computationally efficient algorithm for sparse reinforcement learning.

In Lecture 22, Natalie Collina tells us about a new swap regret algorithm from two brand new papers (Dagan et al. and Peng and Rubinstein) that derive a fundamentally new swap regret algorithm. Its neat! I'll try and summarize. 🧵 youtu.be/58ob2ITmfJk

Exciting work w/ Yuval Dagan Maxwell Fishelson Noah Golowich on efficient algos for no-swap regret learning and, relatedly, correlated eq when the #actions is exponentially large/infinite. While classical works point in the opposite direction, we show that this is actually possible!

Hi everybody - excited to announce the "ITALT" symposium between ITA and ALT that I'm organizing in San Diego together with daniel hsu, Claire Vernade @claireve.bsky.app and Alon Orlitsky to bridge the information theory and learning theory communities!

Another Tuesday, another talk. Noah Golowich (MIT). "Exploring and Learning in Sparse Linear MDPs without Computationally Intractable Oracles". 6pm UTC. Be there or be sparse?

2 concurrent papers, one by Edith Cohen, Jelani Nelson, Lyu, Stemmer and Sarlós (x.com/jalajupadhyay/…) & the other by us solve our open problem presented in #COLT2022. (See Noah Golowich & Livni's upper bound) Interestingly, both have the same hard instance in the lower bound.

🧵1/3: population dynamics are important for the study of epidemiology, i.e. spread of infectious diseases, and controlling hospital flows. Recent work with @Golowich, Zhou Lu, Dhruv Rohatgi and Jennifer Sun, on online control in population dynamics: arxiv.org/abs/2406.01799

We are delighted to announce the recipient of the ACM Doctoral Dissertation Award, 🌱 Nivedita Arora 🌎 Sustainable Computing of Northwestern , and Honorable Mentions, Gabriele Farina of the Massachusetts Institute of Technology (MIT) and William Kuszmaul of Harvard University . Here's to a new generation of computing excellence: bit.ly/3Xm1YyT

Happy to share our work "RL in LMDPs is Tractable:Online Guarantees via OPE" will be presented in #NeurIPS24 🍁🍁🍁 It provides the first assumption-free RL algorithm to LMDPs: a decision problem with M latent and unobserved contexts. w/ Jeongyeol Kwon, Constantine Caramanis, shiemannor

Tomorrow we are hosting Noah ! Please be aware of the time change. The talk is at 5 pm UTC meaning 9am PST, noon EST, 5 pm GMT, 6 pm CET ( to list some example).

Looking forward to presenting this paper on undetectable watermarking of language models at NeurIPS today at 11am, Poster Session 3 East, #4707. Come say hi! arxiv.org/abs/2406.02633