Abstract:In recent years, Deep Reinforcement Learning (DRL) has become a popular paradigm in machine learning due to its successful applications to real-world and complex systems. However, even the state-of-the-art DRL models have been shown to suffer from reliability concerns -- for example, their susceptibility to adversarial inputs, i.e., small and abundant input perturbations that can fool the models into making unpredictable and potentially dangerous decisions. This drawback limits the deployment of DRL systems in safety-critical contexts, where even a small error cannot be tolerated. In this work, we present a comprehensive analysis of the characterization of adversarial inputs, through the lens of formal verification. Specifically, we present the Adversarial Rate, a metric adapted from the ProVe family, for the systematic evaluation of adversarial inputs in DRL, which partitions the input domain into subregions to enable both quantification and spatial visualization of adversarial inputs. The main contribution of this work is to provide a comprehensive evaluation framework for the effect of adversarial inputs on DRL policies. We present a set of tools and algorithms for its computation. Our analysis empirically demonstrates how adversarial inputs can affect the safety of a given DRL system with respect to such perturbations. Moreover, we analyze the behavior of these configurations to suggest several useful practices and guidelines to help mitigate the vulnerability of trained DRL networks.
| Comments: | Accepted to AISoLA 2025 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2402.05284 [cs.LG] |
| (or arXiv:2402.05284v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2402.05284 arXiv-issued DOI via DataCite |
From: Guy Amir [view email]
[v1]
Wed, 7 Feb 2024 21:58:40 UTC (698 KB)
[v2]
Sat, 2 May 2026 21:42:12 UTC (647 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。