Summary of Proposer-agent-evaluator(pae): Autonomous Skill Discovery For Foundation Model Internet Agents, by Yifei Zhou et al.
Proposer-Agent-Evaluator(PAE): Autonomous Skill Discovery For Foundation Model Internet Agents
by Yifei Zhou, Qianlan Yang, Kaixiang Lin, Min Bai, Xiong Zhou, Yu-Xiong Wang, Sergey Levine, Erran Li
First submitted to arxiv on: 17 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Proposer-Agent-Evaluator (PAE) system enables foundation model agents to autonomously discover and practice skills in the wild. This is achieved through a context-aware task proposer that proposes tasks for the agent to practice, with tasks evaluated by an autonomous VLM-based success evaluator serving as the reward signal for refining the agent’s policies via reinforcement learning. The PAE system is validated on challenging vision-based web navigation tasks using real-world and self-hosted websites from WebVoyager and WebArena.To. This work represents the first effective learning system to apply autonomous task proposal with RL for agents that generalizes real-world human-annotated benchmarks with SOTA performances. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine an agent that can navigate the internet or perform tasks in the physical world, like finding directions or buying items online. To make this happen, we need a way for the agent to learn new skills and generalize them to different situations. In this paper, the authors propose a system called Proposer-Agent-Evaluator (PAE) that allows agents to autonomously discover and practice new skills. The PAE system uses a special task proposer that suggests tasks for the agent to try, based on information about the environment. The agent then tries out these tasks and gets feedback in the form of rewards or penalties, which helps it learn and improve its performance. |
Keywords
» Artificial intelligence » Reinforcement learning
