Robust Resilient Signal Reconstruction under Adversarial Attacks

O.M Anubi, L. Mestha, and H. Achanta

• Preprint Manuscript
  
  We consider the problem of signal reconstruction for a system under sparse unbounded signal corruption by an adversarial agent. The reconstruction problem follows the standard error coding problem that has been studied extensively in literature, with the added consideration of support estimation of the attack vector. The problem is formulated as a constrained optimization problem–merging exciting developments in the field of machine learning and estimation theory. Sufficient conditions for the reconstructability and the associated reconstruction error bounds were obtained for both exact and inexact support estimation of the attack vector. Special cases of data-driven model and linear dynamical systems were also considered.