Research map / Adversarial robustness and security / Black-box and transfer attacks
Natural Black-Box Adversarial Examples against Deep Reinforcement Learning
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 feedback-driven advRL-GAN instead of a transferable C&W attack | A3C target policy · Breakout | expected accumulated reward (Black): 44.40 -> 38.23 | Helped |
| use feedback-driven advRL-GAN instead of FGSM with ϵ = .003 | A3C target policy · Pong | expected accumulated reward (Black): 19.50 -> -12.07 | Helped |
| use feedback-driven advRL-GAN instead of a transferable C&W attack | A3C target policy · Chopper Command | expected accumulated reward (Black): 1890.0 -> 1766.7 | Helped |
| add discriminator Dω (ablation) | advRL-GAN · Pong | ΔR: -8.6 | Mixed |
| add discriminator Dω (ablation) | advRL-GAN · Breakout; Chopper Command; Space Invaders; MsPacman; Qbert | ΔR: 6.1; 266.6; -551.0; -526.7; 390.0 | Helped |
| add Ladv loss (ablation) | advRL-GAN · Breakout; Pong; Chopper Command; Space Invaders; MsPacman; Qbert | ΔR: 27.6; 32.6; 83.3; 88.0; 545.3; 670.0 | Helped |
| add Llatent loss (ablation) | advRL-GAN · Breakout; Pong; Chopper Command; Space Invaders; MsPacman; Qbert | ΔR: 2.8; 33.0; 743.3; 177.5; 387.3; 635.0 | Helped |
| add Lnorm loss (ablation) | advRL-GAN · Breakout; Pong; Chopper Command; Space Invaders; MsPacman; Qbert | ΔR: 34.4; 32.2; 350.0; 151.5; 372.3; 570.5 | Helped |
Closest papers by meaning
- Reinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters (AAAI 2025 · similarity 0.88)
- NATTACK: Learning the Distributions of Adversarial Examples for an Improved Black-Box Attack on Deep Neural Networks (ICML 2019 · similarity 0.88)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.88)
- Representation-Based Robustness in Goal-Conditioned Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.85)
- Diffusion Guided Adversarial State Perturbations in Reinforcement Learning (NeurIPS 2025 · 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.