AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations

Huan Zhang, Hongge Chen, Chaowei Xiao, Bo Li, Mingyan Liu, Duane Boning, Cho-Jui Hsieh

NeurIPS 2020 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 38 citations (OpenAlex)

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 changedWhereMeasuredEffect
add state-adversarial regularizer (SA-PPO, SGLD) to PPOPPO · Hopper best attackreward under best attack: 733.8 -> 1403.3Improved
add state-adversarial regularizer (SA-PPO, SGLD) to PPOPPO · Walker2d best attackreward under best attack: 913.7 -> 2468.4Improved
add state-adversarial regularizer (SA-PPO, SGLD) to PPOPPO · Humanoid best attackreward under best attack: 884.1 -> 6073.8Improved
add state-adversarial regularizer to PPOPPO · Humanoid naturalnatural reward: 5270.6 -> 6624.0Improved
naive adversarial training (100% steps) of PPOPPO · Hopper naturalnatural reward: 3167.6 -> 6.1Hurt
use convex relaxation instead of SGLD to solve regularizer in SA-PPO (ablation)SA-PPO · Hopper best attackreward under best attack: 1403.3 -> 1224.2Mixed
add state-adversarial regularizer (SA-DDPG) to DDPGDDPG · Ant best attackreward under best attack: 142 -> 2007Improved
add state-adversarial regularizer (SA-DDPG) to DDPGDDPG · Walker2d naturalnatural reward: 1870 -> 4540Improved
naive adversarial training (50%) of DDPGDDPG · Walker2d best attackreward under best attack: 790 -> 837No clear effect
add SA-DQN regularizer (convex relaxation) to DQNDQN · RoadRunner PGD-10reward under attack: 0.0 -> 44732.0Improved
use convex relaxation instead of PGD to train SA-DQN (ablation)SA-DQN · RoadRunner PGD-10reward under attack: 20402.0 -> 44732.0Improved
adversarial training of DQN (50% frames)DQN · Pong naturalnatural reward: 21.0 -> 10.1Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy