Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Visualizing and Understanding Atari Agents
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 |
|---|---|---|---|
| Add saliency maps to the video | Human judgments of Breakout agent robustness · Breakout survey | Control identified as more robust (%): 35.5 -> 58.1 | Helped |
| Add saliency maps to the video | Human judgments of Breakout agent robustness · Breakout survey | Overfit identified as more robust (%): 48.4 -> 25.8 | Helped |
| Add saliency maps to the video | Human judgments of Breakout agent robustness · Breakout survey | Can't tell (%): 16.1 -> 16.1 | No clear effect |
| Use perturbation-based saliency instead of Jacobian saliency | Atari agent visual explanations · Atari 2600 environments | Interpretability of saliency maps | Helped |
| Add hint pixels encoding the expert's action | Supervised Atari policy agents · Pong, Breakout, and SpaceInvaders | Reliance on hint pixels | Helped |
| Add saliency maps to the video | Human identification of the overfit agent's visual cues · Breakout survey | Respondents identifying hint pixels as the overfit agent's primary input (%): 67.7% | Helped |
Closest papers by meaning
- Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning (ICLR 2020 · similarity 0.92)
- Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Machine versus Human Attention in Deep Reinforcement Learning Tasks (NeurIPS 2021 · similarity 0.89)
- Finding and Visualizing Weaknesses of Deep Reinforcement Learning Agents (ICLR 2020 · similarity 0.89)
- Agent57: Outperforming the Atari Human Benchmark (ICML 2020 · similarity 0.89)
- Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution (ICLR 2020 · similarity 0.89)
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.