notesum.ai

Published at December 3

Defending Against Diverse Attacks in Federated Learning Through Consensus-Based Bi-Level Optimization

cs.LG

cs.CR

cs.MA

math.AP

Released Date: December 3, 2024

Authors: Nicolás García Trillos¹, Aditya Kumar Akash², Sixu Li³, Konstantin Riedl⁴, Yuhua Zhu⁵

Aff.: ¹University of Wisconsin-Madison, Department of Statistics; ²Google; ³University of Wisconsin-Madison; ⁴University of Oxford, Mathematical Institute; ⁵University of California, Los Angeles, Department of Statistics and Data Science

Arxiv: http://arxiv.org/pdf/2412.02535v1

	FedCBO	Oracle Min	Oracle Max	FedCB²O
Overall Acc	$84.26\pm 0.18$	$82.75\pm 0.24$	$\bf{84.43\pm 0.09}$	$82.79\pm 0.14$
Source Class Acc	$40.23\pm 4.02$	$63.53\pm 1.97$	$\bf{63.80\pm 1.76}$	$\bf{55.73\pm 2.94}$
Attack Success Rate	$55.23\pm 3.81$	$\bf{31.08\pm 2.64}$	$32.19\pm 1.93$	$\bf{38.73\pm 3.42}$