Adversarially Robust Multitask Adaptive Control

2025-11-07

AI Summary

This paper addresses collaborative adaptive control of multiple heterogeneous systems under model uncertainty and adversarial data corruption. To counter malicious node interference and system heterogeneity, we propose a unified framework integrating clustering-driven multi-task learning, robust system identification, and attack-resilient aggregation, coupled with a confidence-equivalent control strategy for distributed optimization. Theoretically, under a bounded fraction of malicious nodes, the expected regret decays inversely with the number of honest systems, yielding a non-asymptotic regret bound. Moreover, the clustering mechanism explicitly quantifies how intra-cluster heterogeneity and adversarial behavior degrade LQR performance. Our approach significantly enhances robustness, learning efficiency, and security of distributed control systems, establishing a novel paradigm for trustworthy multi-agent control.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We study adversarially robust multitask adaptive linear quadratic control; a setting where multiple systems collaboratively learn control policies under model uncertainty and adversarial corruption. We propose a clustered multitask approach that integrates clustering and system identification with resilient aggregation to mitigate corrupted model updates. Our analysis characterizes how clustering accuracy, intra-cluster heterogeneity, and adversarial behavior affect the expected regret of certainty-equivalent (CE) control across LQR tasks. We establish non-asymptotic bounds demonstrating that the regret decreases inversely with the number of honest systems per cluster and that this reduction is preserved under a bounded fraction of adversarial systems within each cluster.

Problem

• Robust adaptive control for multiple systems under model uncertainty and adversarial attacks
• Clustered multitask learning with resilient aggregation to mitigate corrupted updates
• Analyzing how clustering accuracy and adversarial behavior affect control regret bounds

Innovation

• Clustered multitask approach for robust control
• Integrates clustering with resilient aggregation
• Mitigates adversarial corruption in model updates