Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Towards Interpretable Reinforcement Learning Using Attention Augmented Agents
Alexander Mott, Daniel Zoran, Mike Chrzanowski, Daan Wierstra, Danilo Jimenez Rezende
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 top-down LSTM-state-dependent queries instead of bottom-up fixed queries or key-norm attention (ablation) | Attention agent · ATARI games in the 7-game comparison | Human normalized score: 541.1% for top-down attention, 274.7% for fixed-query attention, and 274.5% for the L2-Norm Key Agent (median) | Improved |
| Use LSTM-state-dependent queries rather than fixed queries (ablation) | Attention agent · Seaquest | Human normalized score: 1435.9% vs 28.2% | Hurt |
| Use LSTM-state-dependent queries rather than fixed queries (ablation) | Attention agent · Enduro | Human normalized score: 274.8% vs 274.7% | No clear effect |
| Suppress low-weight attention regions with a hard cutoff at t=0.1 (ablation) | Attention agent · Breakout, Ms. Pacman, Seaquest, and Space Invaders | Relative score versus uncut attention: 88 % ± 11 %; 89 % ± 13 %; 116 % ± 29%; 98 % ± 12% | No clear effect |
Closest papers by meaning
- Self-Supervised Attention-Aware Reinforcement Learning (AAAI 2021 · similarity 0.90)
- Focus-Then-Decide: Segmentation-Assisted Reinforcement Learning (AAAI 2024 · similarity 0.87)
- Deep Reinforcement Learning with Stacked Hierarchical Attention for Text-based Games (NeurIPS 2020 · similarity 0.87)
- Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Inverse Attention Agents for Multi-Agent Systems (ICLR 2025 · similarity 0.85)
- Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models (NeurIPS 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.
