🔥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

Cross Domain Imitation Learning deepai.org/publication/cr… by Kun ho Kim et al. including _monste.er_ #ImitationLearning #ComputerScience

New paper on disentanglement c: Given the recent impossibility results in unsupervised disentanglement, we decided to be optimistic and instead provide guarantees (unimpossibility results?) via weak supervision (1/13) arxiv.org/abs/1910.09772 git.io/Je0gN

Some metadata for those curious about their #ICLR2020 reviews. 1. Histogram of the average reviews. 2. Top x% deciles Seems like reviews this year at ICLR 2026 are substantially lower than previous years. Probably an artifact of the new [1,3,6,8] reviewing system. (1/n)

3/3 Tomorrow 9am/11pmPT: Kuno Kim Active Learning with Progress Curiosity. New curiosity signal that is robust, scalable & strong. And leads to emergence of social attention behaviors! paper: arxiv.org/abs/2007.07853 Nick Haber Meg Sano Julian De Freitas

A new scalable and amazingly easy to implement Progress curiosity algorithm released! We report state-of-the-art sample efficiency results in agent-rich 3D environments. We hope to see researchers try it out in various other domains, e.g Atari, Mujoco!

Predicting What You Already Know Helps: Provable Self-Supervised Learning We analyze how predicting parts of the input from other parts (missing patch, missing word, etc.) helps to learn a representation that linearly separates the downstream task. arxiv.org/abs/2008.01064 1/2

"It already does that" video is probably THE video from today so good 🤣 #AppleEvent

This needs to be said out loud, particularly by gatekeepers of AI/ML academia (Profs, advisors, mentors, reviewers, SACs, ACs) Novelty for the sake of it is not a virtue! Many students trying to be different for the sake of novelty and often sacrificing utility in the process!

1/ Def'n: a "learning rule" is a functional that converts error signals (for some given objective function) to changes in system parameters (e.g. synaptic strengths) such that error decreases after iterated application.

Our most recent work is out in Nature! We're reporting on (reinforcement) learning to navigate Loon stratospheric balloons and minimizing the sim2real gap. Results from a 39-day Pacific Ocean experiment show RL keeps its strong lead in real conditions. nature.com/articles/s4158…

Can we make better use of negative samples in contrastive learning? In our #NeurIPS2020 paper, we show this is true by simply using a multi-label objective. Come to our oral presentation at 6:15 PT (neurips.cc/virtual/2020/p…) and poster at 9-11 for more details! Stefano Ermon

🚨 If a provider predicts that a vaccine is 95% effective for you, should you trust the 95% when making decisions? We show how to enable decisions with complete confidence. AISTATS oral happening in 1 hour virtual.aistats.org/virtual/2021/s… Blog ermongroup.github.io/blog/mechanism/