Teacher Algorithms for Deep RL Agents that Generalize in Procedurally Generated Environments – Developmental Systems, a Blog of the Flowers Lab

Developmental Systems, a Blog of the Flowers Lab

TL;DR Teaching algorithms are becoming a key ingredient to scaffold Deep RL agents into their learning odyssey, leading to policies that can generalize to multiple environments. This blog presents this emerging Deep RL sub-field and showcases recent work done in the Flowers Lab on designing a Learning Progress based teacher algorithm. Building Deep RL agents that generalize . Leveraging advances in both training procedures and scalability, the Deep Reinforcement Learning community is now able to tackle ever more challenging control problems. This dynamic is shifting the sets of environments th

saved by

• João Araújo

related reading

• How DeepMind's Generally Capable Agents Were Trained — LessWronglesswrong.com
• Just Ask for Generalization | Eric Jangevjang.com
• Pedagogical RL: Teaching Models to Teach Themselves from Privileged Information - Noah Ziemsnoahziems.com
• State of RL for reasoning LLMs | A. Weersaweers.de
• Learning Beyond Gradientstrinkle23897.github.io
• [AN #159]: Building agents that know how to experiment, by training on procedurally generated games — AI Alignment Forumalignmentforum.org
• Key Papers in Deep RL - Spinning Up documentationspinningup.openai.com
• A Taxonomy of RL Environments for LLM Agentsleehanchung.github.io
• [1709.06560] Deep Reinforcement Learning that Mattersarxiv.org
• Algorithms — Ray 2.55.1docs.ray.io
• Reinforcement Learning via Implicit Imitation Guidancearxiv.org
• PopuLoRA: Co-Evolving LLM Populations for Reasoning Self-⁠Playvmax.ai