Grainger College of Engineering, All Events

Abstract: We present recent advances in using discrete-geometric principles to enable resilient coordination and data aggregation in multiagent networks —from robotic swarms to distributed learning systems—where agents with only partial information must optimize a global objective despite adversarial corruption. In the first part, we develop resilient aggregation schemes for distributed optimization based on data depth, introducing centerpoint-based rules that provably limit malicious influence while maintaining accuracy guarantees. We then move beyond the idealized case of exact benign information to imprecise (set-valued) measurements, including regimes where the imprecision varies over time, and provide aggregation methods with provable guarantees under this dual uncertainty. Next, to address regimes where adversaries exceed classical resilience thresholds, we leverage Tukey medians, characterize when they are strictly preferable to centerpoints, and derive graceful-degradation bounds that quantify how far the aggregate can drift from the correct value beyond the nominal threshold, thereby moving past the traditional “all-or-nothing” resilience paradigm. Finally, we present complementary results on graph distinguishability, giving controllability-based sufficient conditions for a network to admit a unique spectral signature, thereby linking dynamical systems structure to longstanding questions in spectral graph theory.

Bio: Waseem Abbas is an Assistant Professor in the Department of Systems Engineering at the University of Texas at Dallas, with affiliate appointments in the Electrical and Computer Engineering and Mechanical Engineering departments.