TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents

Panagiota Kiourti, Kacper Wardega, Susmit Jha, Wenchao Li
3/1/2019
cs.CRcs.LGstat.ML

Official code: pkiourti/rl_backdoor (20 stars, last commit 2020-07-23).

Abstract

Recent work has identified that classification models implemented as neural networks are vulnerable to data-poisoning and Trojan attacks at training time. In this work, we show that these training-time vulnerabilities extend to deep reinforcement learning (DRL) agents and can be exploited by an adversary with access to the training process. In particular, we focus on Trojan attacks that augment the function of reinforcement learning policies with hidden behaviors. We demonstrate that such attacks can be implemented through minuscule data poisoning (as little as 0.025% of the training data) and in-band reward modification that does not affect the reward on normal inputs. The policies learned with our proposed attack approach perform imperceptibly similar to benign policies but deteriorate drastically when the Trojan is triggered in both targeted and untargeted settings. Furthermore, we show that existing Trojan defense mechanisms for classification tasks are not effective in the reinforcement learning setting.

View on arXivView PDF

Code Implementations(1)

pkiourti/rl_backdoorOfficial100%

Open source implementation of the TrojDRL algorithm presented in TrojDRL: Evaluation of backdoor attacks on Deep Reinforcement Learning

207PythonNov 19, 20186 years ago

Similar Papers

PGN: A perturbation generation network against deep reinforcement learning

88% match2023

Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations

88% match1 repo1442020

SleeperNets: Universal Backdoor Poisoning Attacks Against Reinforcement Learning Agents

87% match2024

BadRL: Sparse Targeted Backdoor Attack Against Reinforcement Learning

87% match1 repo92023

Mitigating Deep Reinforcement Learning Backdoors in the Neural Activation Space

87% match2024

Beyond Training-time Poisoning: Component-level and Post-training Backdoors in Deep Reinforcement Learning

86% match2025

Cite this paper

@article{kiourti2019trojdrl,
  title  = {TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents},
  author = {Panagiota Kiourti and Kacper Wardega and Susmit Jha and Wenchao Li},
  year   = {2019},
  eprint = {1903.06638},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1903.06638}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1903.06638.svg)](https://codeofpaper.com/paper/1903.06638)

Discussion