Research map / Adversarial robustness and security / Attacks on text and graphs
Online Robust Policy Learning in the Presence of Unknown Adversaries
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 separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=1.0/-) · InvertedPendulum-v2 (m/n=1.0/-) | normalized avg. training return: 0.96 ± 0.03 -> 0.96 ± 0.03 | No clear effect |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.995/0.005) · InvertedPendulum-v2 (m/n=0.995/0.005) | normalized avg. training return: 0.238 ± .082 -> 0.553 ± 0.242 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.995/0.005) · InvertedPendulum-v2 (m/n=0.995/0.005) | normalized avg. evaluation return: 0.471 ± 0.051 -> 0.99 ± 0.001 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.95/0.05) · InvertedPendulum-v2 (m/n=0.95/0.05) | normalized avg. training return: 0.612 ± .08 -> 0.677 ± 0.149 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.95/0.05) · InvertedPendulum-v2 (m/n=0.95/0.05) | normalized avg. evaluation return: 0.644 ± 0.078 -> 0.99 ± 0.001 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.8/0.2) · InvertedPendulum-v2 (m/n=0.8/0.2) | normalized avg. training return: 0.613 ± 0.043 -> 0.728 ± 0.063 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.8/0.2) · InvertedPendulum-v2 (m/n=0.8/0.2) | normalized avg. evaluation return: 0.539 ± 0.023 -> 0.994 ± 0.165 | Helped |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.5/0.5) · InvertedPendulum-v2 (m/n=0.5/0.5) | normalized avg. training return: 0.749 ± 0.093 -> 0.764 ± 0.078 | No clear effect |
| Use separate nominal and adversarial policies instead of one Vanilla PPO policy | InvertedPendulum-v2 (m/n=0.5/0.5) · InvertedPendulum-v2 (m/n=0.5/0.5) | normalized avg. evaluation return: 0.787 ± 0.010 -> 0.948 ± 0.086 | Helped |
| Use oracle attack-condition selection rather than advantage-based master detection | MLAH on Hopper-v2 · Hopper-v2 | average performance during attack | Hurt |
Closest papers by meaning
- Beyond Worst-case Attacks: Robust RL with Adaptive Defense via Non-dominated Policies (ICLR 2024 · similarity 0.88)
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Off-Policy Actor-Critic for Adversarial Observation Robustness: Virtual Alternative Training via Symmetric Policy Evaluation (ICML 2025 · similarity 0.88)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.87)
- Multi-level Certified Defense Against Poisoning Attacks in Offline Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Robust and Diverse Multi-Agent Learning via Rational Policy Gradient (NeurIPS 2025 · similarity 0.86)
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.