Advantage-based Temporal Attack in Reinforcement Learning

TL;DR

This paper introduces AAT, a novel adversarial attack method in reinforcement learning that leverages a multi-scale causal self-attention mechanism and advantage weighting to generate more temporally correlated and effective adversarial examples, outperforming existing methods.

Contribution

The paper proposes AAT, a new advantage-based adversarial attack method that captures temporal dependencies and guides perturbation generation using a weighted advantage mechanism.

Findings

AAT outperforms existing attack baselines on Atari, DeepMind Control Suite, and Google Football tasks.

AAT generates more temporally correlated adversarial examples, improving attack success.

AAT effectively leverages historical information and advantage weighting for stronger attacks.

Abstract

Extensive research demonstrates that Deep Reinforcement Learning (DRL) models are susceptible to adversarially constructed inputs (i.e., adversarial examples), which can mislead the agent to take suboptimal or unsafe actions. Recent methods improve attack effectiveness by leveraging future rewards to guide adversarial perturbation generation over sequential time steps (i.e., reward-based attacks). However, these methods are unable to capture dependencies between different time steps in the perturbation generation process, resulting in a weak temporal correlation between the current perturbation and previous perturbations.In this paper, we propose a novel method called Advantage-based Adversarial Transformer (AAT), which can generate adversarial examples with stronger temporal correlations (i.e., time-correlated adversarial examples) to improve the attack performance. AAT employs a multi-scale causal self-attention (MSCSA) mechanism to dynamically capture dependencies between historical information from different time periods and the current state, thus enhancing the correlation between the current perturbation and the previous perturbation. Moreover, AAT introduces a weighted advantage mechanism, which quantifies the effectiveness of a perturbation in a given state and guides the generation process toward high-performance adversarial examples by sampling high-advantage regions. Extensive experiments demonstrate that the performance of AAT matches or surpasses mainstream adversarial attack baselines on Atari, DeepMind Control Suite and Google football tasks.