Research map / Reinforcement learning / Visual and representation RL
ProtoX: Explaining a Reinforcement Learning Agent via Prototyping
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use ProtoX instead of VIPER for image-state policy imitation | PPO expert policy · Pong | accuracy: 45% -> 95% | Helped |
| Use ProtoX instead of VIPER for image-state policy imitation | PPO expert policy · Seaquest | accuracy: 31% -> 96% | Helped |
| Use ProtoX instead of GAIfO for policy imitation | PPO expert policy · Pong | accuracy: 19% -> 95% | Helped |
| Use ProtoX instead of GAIfO for policy imitation | PPO expert policy · Seaquest | accuracy: 11% -> 96% | Helped |
| Use ProtoX instead of GAIfO for policy imitation | PPO expert policy · Mario 1-1 | accuracy: 33% -> 93% | Helped |
| Use ProtoX instead of GAIfO for policy imitation | PPO expert policy · Mario 8-3 | accuracy: 25% -> 95% | Helped |
| Use interpretable prototypes rather than a ResNet-BC black-box classifier | Behavior-cloned PPO expert policy · Pong, Seaquest, Mario 1-1, and Mario 8-3 | accuracy: 3% drop on average | Hurt |
| Use ProtoX instead of VIPER to reproduce action changes | PPO expert policy · Pong | flip-point sensitivity: 44% -> 84% | Helped |
| Use ProtoX instead of VIPER to reproduce action changes | PPO expert policy · Seaquest | flip-point sensitivity: 24% -> 87% | Helped |
| Use ProtoX instead of GAIfO to reproduce action changes | PPO expert policy · Pong | flip-point sensitivity: 22% -> 84% | Helped |
| Use ProtoX instead of GAIfO to reproduce action changes | PPO expert policy · Seaquest | flip-point sensitivity: 6% -> 87% | Helped |
| Use ProtoX instead of GAIfO to reproduce action changes | PPO expert policy · Mario 1-1 | flip-point sensitivity: 18% -> 94% | Helped |
Closest papers by meaning
- Reinforcement Learning with Prototypical Representations (ICML 2021 · similarity 0.89)
- This State Looks Like That: Self-Interpretable Reinforcement Learning Agents using Prototype Soft Actor-Critic (ICML 2026 · similarity 0.89)
- ProTo: Program-Guided Transformer for Program-Guided Tasks (NeurIPS 2021 · similarity 0.87)
- Explaining Reinforcement Learning Agents through Counterfactual Action Outcomes (AAAI 2024 · similarity 0.86)
- VLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought (NeurIPS 2024 · similarity 0.85)
- Agent-Aware Training for Agent-Agnostic Action Advising in Deep Reinforcement Learning (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.