notesum.ai

Published at December 10

Defending Against Neural Network Model Inversion Attacks via Data Poisoning

cs.CR

Released Date: December 10, 2024

Authors: Shuai Zhou¹, Dayong Ye², Tianqing Zhu, Wanlei Zhou

Aff.: ¹Faculty of Data Science, City University of Macau, Macau, China; ²Centre of Cyber Security and Privacy and the School of Computer Science, University of Technology Sydney, Australia

Arxiv: http://arxiv.org/pdf/2412.07575v1

Atta.	Defe.	Utility		Privacy
loss	loss	Model Acc.	Conf. Dist.	Attack Acc.	Recon. Error
MSE	No-defense	$94.82\%$	-	$32.32\%$	$0.2076$
	MSE	$94.82\%$	0.00340	$5.08\%$	$0.2318$
	PSNR	$94.82\%$	$0.00363$	$5.04\%$	$0.2335$
PSNR	No-defense	$94.82\%$	-	$31.17\%$	$0.2104$
	MSE	$94.82\%$	0.00340	4.06%	0.2321
	PSNR	$94.82\%$	$0.00363$	$4.63\%$	$0.2316$