RT @luisa_zintgraf: 🎉 Our Meta-RL survey is now published in Foundations and Trends in Machine Learning! A deep dive into how agents can le….

@luisa_zintgraf @chelseabfinn @shimon8282 @ristovuorio For more, see our podcast interview.( . and conference presentation (.

4 years, 120 pages (and only a couple trade wars) later, it’s finally here. Huge thanks to my brilliant co-authors — Evan Liu, Zheng Xiong, @luisa_zintgraf, @chelseabfinn, @shimon8282, and especially @ristovuorio. Published Paper (with sample text).

Big news—our survey paper “A Tutorial on Meta-Reinforcement Learning” is officially published!. Meta-RL = learning how to adapt through interaction. It embraces The Bitter Lesson: don’t hardcode agents—train them to adapt on their own. 🧵⬇️.

2️⃣ Non-Markovian Feedback is Crucial: VLMs lack action-conditioned data, so visual cues over time are needed to assess progress. 3️⃣ Simplicity Outperforms Complexity: A filtered and weighted behavior cloning approach surpasses complex RL-based methods. 🔗

Still, VLMs can recognize successful task completion and provide valuable feedback to guide the learning of RL agents. 🔑 Key insights.1️⃣ Sub-Trajectories Matter: Full-trajectory preference learning worsens stitching issues, so sub-sampling trajectories is critical

✈️ SFO: Piloting VLM Feedback for Offline RL. I’m excited to share a preliminary study on VLMs for offline reinforcement learning!. A lack of internet-scale action data has prevented foundation models from natively understanding control. 🧵.

RT @ShangtongZhang: Excited to share our new survey of in-context reinforcement learning!! w/ @AmirMoeini99 @wangji….

Many thanks to @instadeepai and my amazing co-authors, @ShikhaSurana01, Manus McAuliffe, Oliver Bent, @tomdbarrett, @juanjogarau, and @pduckw, for their collaboration and support!.

🗓️ Don’t miss it: Sunday, 3-3:30 PM, West Meeting Room 202-204!. 📄 Read the paper: 🔗 Check out the project: 🎥 Watch the award-winning video:

🎉🚨 Big news! Our research, Metalic: Meta-Learning In-Context with Protein Language Models, 🧬 won a competition! #NeurIPS2024🤖📚. We advance in-context learning and protein fitness prediction with this paradigm:.✨ Pre-training.🔥 Learning to in-context learn🔥.✨ Fine-tuning

Many thanks to the organizers of the @AutoRL_Workshop for the lively discussion. Check out the workshop page: 🔗.

Missed this provocative panel? I was honored to share the stage at #ICML2024 with @pcastr, @XingyouSong, and my colleague @AlexDGoldie! We discussed future perspectives on automated RL, meta-learning, and LLMs 🤖. Catch the discussion here: at 7:16:00 🎙️

Wouldn’t it be nice if ChatGPT could figure out how to do your taxes?🧾🤖This would be in-context RL. learning to do in-context-RL is actually the problem studied in meta-RL. Come to our tutorial at 4:15 today in room 119 @ #AAAI24 to learn more!.

RT @whi_rl: 🤖 Come chat with @jakeABeck TODAY about his outstanding work on meta-RL, memory, and hypernetworks!🤖. ⏰ This morning at 10:45 a….

At NeurIPS this week presenting Recurrent Hypernetworks are Surprisingly Strong in Meta-RL 🔗 👨‍🏫🤖. Come chat about Meta-RL, Memory, or Hypernetworks! 🍻.

RT @whi_rl: RL agents are notoriously slow to learn 🐢. However, meta-RL can make RL agents that learn fast!🔥. Check out this talk introduci….

Heyyo! Was just interviewed by the TalkRL Podcast!! 🎙️🔥. @ristovuorio and I explain meta-RL. Give it a listen!.

With promising applications in sight and an array of open problems, we expect meta-RL research to continue to boom!.

Still, there are many open problems, such as out-of-distribution (OOD) generalization and learning over a broad distribution of tasks.