Revealing the Risk of Hyper-Parameter Leakage in Deep Reinforcement Learning Models

Summary

Researchers show that the hyper-parameters of a closed-box deep reinforcement learning model can be inferred from its observable states and actions. They introduce a hyper-parameter inference attack named HyperInfer, which uses two state generation methods that provoke divergent model responses. Inference accuracy exceeds 90% in some settings, such as PPO with CartPole.

Mitigation

The source proposes potential defensive strategies to minimize the risk of hyper-parameter leakage in DRL models but does not specify them in the provided text.

Related items

• InfoModel Extraction of Convolutional Neural Networks with Max-PoolingSimilar attack · OpenAlex (peer-reviewed AI security)
• InfoTrustworthy Distributed Machine Learning in Cloud-Edge-IoT Systems: A Heterogeneity-Aware Survey of Security, Privacy, and RobustnessSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoPreventing Model Extraction in Contextual Linear OptimizationSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoMapSteal: Map-based Autonomous Knowledge Stealing Attack Against Large Language ModelsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoHow to Secure AI Models in Production Environments | Endor LabsSimilar attack · Endor Labs Blog