If you are doing offline RL, and you have a bunch of data without reward labels, should you: (a) learn a reward function; (b) label all the data with 0? Turns out that (b) (somewhat shockingly) is a very good choice, in theory and in practice: arxiv.org/abs/2202.01741 A thread:

CPG-RL: Learning Central Pattern Generators for Quadruped Locomotion abs: arxiv.org/abs/2211.00458

I am extremely happy to announce the mamba 1.0 release 🎉 If you are curious on what's new, check out this blog post: medium.com/@wolfv/releasi… Thanks to all contributors, users (& their feedback) we can present the most stable and fastest mamba ever 🚀

Training #ReinforcementLearning algorithms from scratch is computationally intensive and time consuming. We propose an alternate approach, Reincarnating RL, that integrates prior computation into the RL training workflow. Learn more and grab the code at goo.gle/3Ws2TLk

Discrete Factorial Representations as an Abstraction for Goal Conditioned RL Riashat Islam, Hongyu Zang, Alex Lamb, Kenji Kawaguchi, Xin Li, Romain Laroche, Yoshua Bengio arxiv.org/abs/2211.00247 NeurIPS'22

Want your robot to learn agile skills without reward shaping or designing expert controllers? We propose WASABI which allows Solo, a quadruped robot to obtain highly dynamic skills (e.g. backflip) from only rough, partial, hand-held human demonstrations. sites.google.com/view/corl2022-…

You couldn't make it up.

90 Days of Diffusion 90 🤯 AI Advances 👇

RL with KL penalties – a powerful approach to aligning language models with human preferences – is better seen as Bayesian inference. A thread about our paper (with Ethan Perez and Chris L Buckley) to be presented at #emnlp2022 🧵arxiv.org/pdf/2205.11275… 1/11

A recently developed machine learning model has given scientists a new schema for predicting the scent of individual molecules: metabolism. quantamagazine.org/ai-model-links…

Team stiff_finger (Pablo Tano Alberto Chiappa Nisheet Patel) receiving their award plaque and part of the organization team (Vittorio Caggiano Guillaume Durandau Vikash Kumar ) during #NeurIPS2022

Congrats to the team! So proud of PhD students Alberto Chiappa, Pablo Tano and Nisheet Patel and co-Alex: Alexandre Pouget

🔥MyoSuite 1.4 released 🔥 sites.google.com/view/myosuite ➡️Validated upper extremity models with interacting exo-robots ➡️4000x faster-than-SOTA (suitable for RL) ➡️Full contact dynamics Extremely excited about release that took >1 year of dev & testing, and our growing community

MyoChallenge '22—a retrospective on progress, lessons learned & key takeaways 🫴🎲 Check it out on our new Medium account: 📽️Videos of the winning policies 🏆 Photos with winners from NeurIPS '22 🧑‍🏫Link to MyoSymposium speakers, talk medium.com/@myosuite/myoc…

We had great fun participating & writing the paper with all the winning teams and organizers-- check it out! I'm quite excited what one can learn about biological motor control based on the new simulators! Our team: Alberto Chiappa Pablo Tano Nisheet Patel Alexandre Pouget

Acquiring musculoskeletal skills with curriculum-based reinforcement learning biorxiv.org/cgi/content/sh… #biorxiv_neursci

Are you interested in motor skills, musculoskeletal control and reinforcement learning? Check out our manuscript: biorxiv.org/content/10.110…

We're super excited to launch Axy at the European Federation for Primatology's #EFP2024 conference 🦍🐵🦧🐒

How does the brain control the numerous muscles of the body? Let's say you want to rotate two balls in your hand, how does your brain achieve that? Read our article in Neuron to learn more! cell.com/neuron/fulltex…

🚀 Axy is the official AI-companion app (in beta) for #COSYNE2025! 🧠 Discover relevant posters, 👥 find poster buddies, and ⏰ never miss important talks with our AI-powered recommendations. 🔍 Made by scientists, for scientists with ❤️. Try it now: 🔗 beta.axy-app.com/cosyne2025?ref… ✨