Poisoning Resilient Federated Learning Under Privacy Constraints
Abstract
Federated learning enables multiple clients to collaboratively train models without exchanging raw data; however, the system remains susceptible to adversarial attacks through maliciously crafted model updates. This challenge intensifies as privacy-preserving methods, including identity obfuscation, attribute randomization, synthetic data generation, encrypted aggregation, and secure computation, restrict the ability to discern individual participant contributions. This chapter proposes methods that identify and control poisoning while complying with strict privacy limitations. It describes the problem, realistic threat and privacy models, and a four-layer defense structure including input sanitation, update filtering, privacy-sensitive robust aggregation, and post-aggregation audits. An experimental plan tests effectiveness in heterogeneous data and high-level privacy environments. The chapter fills a significant gap by reconciling privacy and robustness and will be valuable to graduate students and researchers studying secure federated learning.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.