.png){kind=icon}
.png){kind=icon}
Return
Resilient distributed online nonconvex optimization algorithm against randomly corrupted attacks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jfranklin.2026.108688.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper focuses on a distributed nonconvex optimization problem with multiple coupled constraints over time-varying (TV) unbalanced directed graphs. The objective of agents in digraphs is to cooperatively with neighbors handle the problem, where local cost functions may be nonconvex and local constraint functions are convex. Furthermore, the consecutive information interactions among agents provides a possible risk that the adversary launches randomly corrupted attacks where gradient information is likely to be replaced by arbitrary information despite of predefined iteration rules. To effectively tackle these issues, we propose a primal dual robust double aggregation (PDRDA) algorithm which introduces temporal and spatial aggregation schemes for primal and dual variables, respectively. Specifically, we utilize an adaptive combination scheme (ACS) to fuse the current clipped gradient and normalized historical gradient used for updating primal variables. Meanwhile, the normalized cumulatively spatial gradient from neighbors and current clipped gradient are also combined with ACS to update dual variables. Besides, PDRDA adopts auxiliary variables to overcome the imbalance induced by digraphs. We present rigorous theoretical analyses to prove that PDRDA sublinearly converges to stationary points and obtains sublinear constraint violation. In numerical simulations, a target localization example is used to verify the validity of the proposed algorithm.
Keywords:
distributed optimization
nonconvex
resilient algorithm
randomly corrupted attacks
constraint violation
Journal
J
IF:
4.2
Papers:
822
Citations:
0