Research map / Reinforcement learning / Visual and representation RL
Deep Reinforcement Learning Policies Learn Shared Adversarial Features across MDPs
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 |
|---|---|---|---|
| Apply a high-sensitivity direction from another MDP using ENR rather than a random perturbation | DDQN policies · Arcade Learning Environment, including BankHeist, RoadRunner, JamesBond, CrazyClimber, TimePilot, and Pong | impact: ENR: 0.496$\pm$0.055, 0.816$\pm$0.034, 0.910$\pm$0.051, 0.993$\pm$0.002, 0.312$\pm$0.118, 0.946$\pm$0.017; Gaussian: 0.0 | Helped |
| Use ENR rather than C&W to compute transferred high-sensitivity directions | DDQN policies · Arcade Learning Environment, cross-MDP setting | impact: BankHeist 0.496$\pm$0.055 vs 0.022$\pm$0.0349; RoadRunner 0.816$\pm$0.034 vs 0.919$\pm$0.018; JamesBond 0.910$\pm$0.051 | Mixed |
| Transfer directions from vanilla DDQN to adversarially trained SA-DDQN policies | SA-DDQN · RoadRunner, Pong, and BankHeist | impact: 0.397$\pm$0.024, 1.0$\pm$0.000, 0.758$\pm$0.042; Gaussian: 0.023$\pm$0.058, 0.019$\pm$0.007, 0.061$\pm$0.012 | Helped |
| Transfer directions from vanilla DDQN in another MDP to adversarially trained SA-DDQN | SA-DDQN · RoadRunner, Pong, and BankHeist | impact: 0.546$\pm$0.014, 0.659$\pm$0.069, 0.241$\pm$0.009 | Mixed |
| Transfer directions across DDQN architectural variants instead of using the same architecture | DDQN variants · Pong | impact: 1.0; environment-independent values: 0.995, 0.985, 0.966, 0.981 | No clear effect |
Closest papers by meaning
- Understanding and Diagnosing Deep Reinforcement Learning (ICML 2024 · similarity 0.92)
- Stable Gradients for Stable Learning at Scale in Deep Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Confounding Robust Deep Reinforcement Learning: A Causal Approach (NeurIPS 2025 · similarity 0.86)
- Geometry of Neural Reinforcement Learning in Continuous State and Action Spaces (ICLR 2025 · similarity 0.86)
- A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 2024 · similarity 0.85)
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.