Federated learning systems that jointly preserve Byzantine robustness and privacy have remained an open problem. Robust aggregation, the standard defense for Byzantine attacks, generally requires server access to individual updates or nonlinear computation -thus is incompatible with privacy-preserving methods such as secure aggregation via multiparty computation. To this end, we propose SHARE (Secure Hierarchical Robust Aggregation), a distributed learning framework designed to cryptographically preserve client update privacy and robustness to Byzantine adversaries simultaneously. The key idea is to incorporate secure averaging among randomly clustered clients before filtering malicious updates through robust aggregation. Experiments show that SHARE has similar robustness guarantees as existing techniques while enhancing privacy.Preprint. Under review.

No abstract

No abstract

No abstract

Tactical selection of experiments to estimate an underlying model is an innate task across various fields. Since each experiment has costs associated with it, selecting statistically significant experiments becomes necessary. Classic linear experimental design deals with experiment selection so as to minimize (functions of) variance in estimation of regression parameter. Typically, standard algorithms for solving this problem assume that data associated with each experiment is fully known. This isn't often true since missing data is a common problem. For instance, remote sensors often miss data due to poor connection. Hence experiment selection under such scenarios is a widespread but challenging task. Though decoupling the tasks and using standard data imputation methods like matrix completion followed by experiment selection might seem a way forward, they perform sub-optimally since the tasks are naturally interdependent. Standard design of experiments is an NP hard problem, and the additional objective of imputing for missing data amplifies the computational complexity. In this paper, we propose a maximum-entropy-principle based framework that simultaneously addresses the problem of design of experiments as well as the imputation of missing data. Our algorithm exploits homotopy from a suitably chosen convex function to the non-convex cost function; hence avoiding poor local minima. Further, our proposed framework is flexible to incorporate additional application specific constraints. Simulations on various datasets show improvement in the cost value by over 60% in comparison to benchmark algorithms applied sequentially to the imputation and experiment selection problems.