Couldn't be more excited! Mark your calendar and stay tuned for the details!

👀 ReDo #ICML2023 #RL

Time for celebrating🎉 One paper accepted for an oral presentation at #ICML2023! I was very lucky to work on this project with amazing collaborators utku , Pablo Samuel Castro , Rishabh Agarwal ! Details soon✨ See you in Hawaii🏝️

Here we go🌟 Check out the details of our #ICML2023 paper!

Don't miss Bram Grooten 's presentation at #ICLR2023 this Friday, where we will be sharing our work on how to deal with noisy environments in RL.

Sparsity workshop stickers in the colors of the Rwandan flag by utku.

Consider supporting researchers in Africa!

Looking for nice supervisor and an interesting project? Check out this opportunity!

Looking forward to presenting our full-paper at @aamas2023 on Friday at 14:00 in the #DeepLearning session. If you're here, come check it out at poster 34 🙂 📜arxiv.org/abs/2302.06548

2. The Dormant Neuron Phenomenon in Deep RL (led by Ghada Sokar w/t utku and Pablo Samuel Castro). Reviewers loved this work and it was accepted as an oral. x.com/pcastr/status/…

📢CLAI Unconf pre-registration paper proposals are due soon (July 21)!📢 Have a scientific hypothesis? Ready to publish in PMLR? Chase SCIENCE instead of SOTA by submitting your idea with us! It is not too late to get started! 👇 unconf.continualai.org/call-for-papers

with utku and Rishabh Agarwal i'll be presenting work led by Ghada Sokar on the dormant neuron phenomenon in deep RL. utku & i give oral on tue at 5:30pm in Ballroom C then with rishabh we'll present the poster in sess 3, wed at 11am, Exhibit Hall 1 #709 x.com/pcastr/status/…

Deep learning may be hard, but deep un-learning is even harder. 💪 How do we efficiently remove the influence of specific training examples while maintaining good performance on the remainder? Announcing NeurIPS Unlearning Competition 📢 Submit your best ideas!🏆

The excitement knows no limits🎉 🚨Only 1 hour to go until the start of CLAI Unconf, but you can still continue joining! 👋Come be part of one of the largest continual learning gatherings with over 400 participants! It's all virtual - it's all free: unconf.continualai.org/register

Google Deepmind presents Mixtures of Experts Unlock Parameter Scaling for Deep RL paper page: huggingface.co/papers/2402.08… The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally

📢Mixtures of Experts unlock parameter scaling for deep RL! Adding MoEs, and in particular Soft MoEs, to value-based deep RL agents results in more parameter-scalable models. Performance keeps increasing as we increase number of experts (green line below)! 1/9

Happening now! Come say hi #ICML2024

📢Don't flatten, tokenize!📢 tl;dr: the key reason for softmoe's efficacy in deep RL turns out to be tokenization! i.e. the common practice of flattening the output of conv encoder layers is quite suboptimal! 👇🏾more details in thread below👇🏾 1/11