Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Deep Reinforcement Learning through Adversarial Loss
Tuomas Oikarinen, Wang Zhang, Alexandre Megretski, Luca Daniel, Tsui-Wei Weng
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 Approach 2 overlap loss instead of Approach 1 upper-bound loss (ablation) | RADIAL-DQN · Atari RoadRunner PGD 1/255 | reward: 4240 -> 44445 | Improved |
| use Approach 2 instead of Approach 1 (ablation) | RADIAL-DQN · Atari Pong PGD 5/255 | reward: 1.1 -> 21.0 | Improved |
| use Approach 2 instead of Approach 1 (ablation) | RADIAL-DQN · Atari BankHeist PGD 5/255 | reward: 1063.5 -> 1182.5 | Improved |
| margin coefficient c=0.5 instead of 0.25 (ablation) | RADIAL-DQN · Atari RoadRunner PGD 5/255 | reward: 11220 -> 23820 | Improved |
| margin coefficient c=0.5 instead of 0.75 (ablation) | RADIAL-DQN · Atari RoadRunner PGD 5/255 | reward: 4870 -> 23820 | Improved |
| symmetric overlap loss instead of default (ablation) | RADIAL-DQN · Atari RoadRunner/BankHeist PGD 5/255 | reward: RoadRunner 23820 -> 21385; BankHeist 1182.5 -> 1305 | Mixed |
| CROWN-IBP bounds instead of IBP (ablation) | RADIAL-DQN · Atari PGD 5/255 | reward: RoadRunner 23820 -> 18315; BankHeist 1182.5 -> 1328.5 | Hurt |
| apply RADIAL training (Approach 2) to DQN versus standard DQN (ablation) | DQN · Atari RoadRunner PGD 1/255 | reward: 0.0 -> 44445 | Improved |
| RADIAL-DQN versus SA-DQN | DQN · Atari RoadRunner PGD 5/255 | reward: 3350 -> 23820 | Improved |
| RADIAL-PPO versus PPO | PPO · ProcGen Fruitbot Eval PGD 5/255 | reward: 8.51 -> 25.81 | Improved |
| RADIAL-PPO versus SA-PPO | PPO · MuJoCo Walker2D MAD 0.15 | reward: 3652.1 -> 3895.3 | Improved |
| RADIAL-PPO versus SA-PPO | PPO · MuJoCo HalfCheetah MAD 0.15 | reward: 4173.1 -> 4008.5 | Mixed |
Closest papers by meaning
- Beyond Worst-case Attacks: Robust RL with Adaptive Defense via Non-dominated Policies (ICLR 2024 · similarity 0.91)
- Breaking the Barrier: Enhanced Utility and Robustness in Smoothed DRL Agents (ICML 2024 · similarity 0.90)
- Representation-Based Robustness in Goal-Conditioned Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Robust Adversarial Reinforcement Learning via Bounded Rationality Curricula (ICLR 2024 · similarity 0.89)
- Enhancing Robustness in Deep Reinforcement Learning: A Lyapunov Exponent Approach (NeurIPS 2024 · similarity 0.89)
- Towards Optimal Adversarial Robust Q-learning with Bellman Infinity-error (ICML 2024 · 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.
