CLAIRE-Labo/no-representation-no-trust

Codebase to fully reproduce the results of "No Representation, No Trust: Connecting Representation, Collapse, and Trust Issues in PPO" (Moalla et al. 2024). Uses TorchRL and provides extensive tools for studying representation dynamics in policy optimization.

32
/ 100
Emerging

This project provides the tools and codebase to understand why Reinforcement Learning (RL) agents, specifically those using Proximal Policy Optimization (PPO), sometimes fail to learn effectively and experience performance collapse. It helps RL researchers analyze how the internal 'representation' of an agent changes over time. You input experimental parameters for PPO agents in environments like Atari or MuJoCo, and it outputs detailed logs and plots showing agent performance and internal representation dynamics, helping you diagnose and mitigate learning failures.

No commits in the last 6 months.

Use this if you are an RL researcher or practitioner investigating stability and plasticity issues in PPO-based agents and need to analyze the underlying causes of performance collapse related to internal representations.

Not ideal if you are looking for a general-purpose RL library to train agents without a specific focus on analyzing representation dynamics and learning stability.

Reinforcement-Learning Policy-Optimization Agent-Stability Deep-RL-Research Learning-Dynamics
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 16 / 25
Community 9 / 25

How are scores calculated?

Stars

31

Forks

3

Language

Python

License

MIT

Last pushed

Nov 20, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/CLAIRE-Labo/no-representation-no-trust"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.