Enhance membership inference attacks in federated learning

Article Computer Science, Theory & Methods

A Comprehensive Survey on Poisoning Attacks and Countermeasures in Machine Learning

Zhiyi Tian et al.

Summary: The prosperity of machine learning has led to an increase in attacks on the training process, with poisoning attacks emerging as a significant threat. Defending against these attacks is challenging, and a systematic review from a unified perspective is lacking. This survey provides a comprehensive overview of poisoning attacks and countermeasures in both centralized and federated learning, categorizing attack methods based on goals and analyzing their differences and connections. Countermeasures in different learning frameworks are presented, along with a discussion of the feasibility of poisoning attacks and potential research directions.

ACM COMPUTING SURVEYS (2023)

Add to Collection

Article Computer Science, Information Systems

Data Poisoning Attacks on Federated Machine Learning

Gan Sun et al.

Summary: This article explores the vulnerability of federated machine learning models, particularly in IoT systems. By attacking a popular federated multitask learning framework, the authors propose a novel systems-aware optimization method to compute gradients for poisoned data and derive optimal attack strategies in federated machine learning. Experiments demonstrate the sensitivity of the federated multitask learning model to both direct and indirect poisoning attacks.

IEEE INTERNET OF THINGS JOURNAL (2022)

Add to Collection

Article Engineering, Electrical & Electronic