Robust Aggregation Rule

Robust aggregation rules (AGRs) are crucial for secure and reliable distributed machine learning, particularly in federated learning settings where data is decentralized and potentially compromised by malicious actors or non-identical data distributions. Current research focuses on developing AGRs that are resilient to Byzantine attacks (malicious data injection) and handle non-identically and independently distributed (non-IID) data, often employing techniques like gradient splitting or distance-based outlier suppression to identify and mitigate the influence of faulty updates. These advancements are vital for ensuring the accuracy and trustworthiness of models trained on sensitive data across multiple sources, with applications ranging from medical imaging to other privacy-sensitive domains.

Papers

February 13, 2023

Byzantine-Robust Learning on Heterogeneous Data via Gradient Splitting
Yuchen Liu, Chen Chen, Lingjuan Lyu, Fangzhao Wu, Sai Wu, Gang Chen
Heterogeneous Data Byzantine Robust Byzantine Attack Robust Aggregation Rule

February 9, 2023

On the Privacy-Robustness-Utility Trilemma in Distributed Learning
Youssef Allouah, Rachid Guerraoui, Nirupam Gupta, Rafael Pinot, John Stephan
Native Robustness Differential Privacy Adversarial Learning Distributed Learning Utility Privacy Trade Robust Aggregation Rule

August 27, 2022

BOBA: Byzantine-Robust Federated Learning with Label Skewness
Wenxuan Bao, Jun Wu, Jingrui He
Native Robustness Label Distribution Byzantine Attack Byzantine Robust Federated Learning Robust Aggregation Rule

July 15, 2022

Suppressing Poisoning Attacks on Federated Learning for Medical Imaging
Naif Alkhunaizi, Dmitry Kamzolov, Martin Takáč, Karthik Nandakumar
Medical Imaging Poisoning Attack Outlier Detection Robust Aggregation Rule

Robust Aggregation Rule

Papers

Byzantine-Robust Learning on Heterogeneous Data via Gradient Splitting

On the Privacy-Robustness-Utility Trilemma in Distributed Learning

BOBA: Byzantine-Robust Federated Learning with Label Skewness

Suppressing Poisoning Attacks on Federated Learning for Medical Imaging