Aggregation Defense

Aggregation defense in federated learning aims to protect collaboratively trained models from malicious data poisoning and backdoor attacks, ensuring the integrity and reliability of the final model. Current research focuses on developing robust aggregation schemes, such as deep partition aggregation, and analyzing their efficiency and effectiveness against increasingly sophisticated attacks, including those leveraging generative models and reinforcement learning. Understanding the practical limitations and vulnerabilities of these defenses is crucial for building secure and trustworthy federated learning systems, impacting the development of privacy-preserving machine learning applications.

Papers

March 10, 2024

Fake or Compromised? Making Sense of Malicious Clients in Federated Learning
Hamid Mozaffari, Sunav Choudhary, Amir Houmansadr
Federated Learning Fake Speech Multiple Sens Malicious Client Adversary Model Aggregation Defense

October 17, 2023

Adversarial Robustness Unhardening via Backdoor Attacks in Federated Learning
Taejin Kim, Jiarui Li, Shubhranshu Singh, Nikhil Madaan, Carlee Joe-Wong
Adversarial Training Backdoor Attack Decentralized Training Aggregation Defense

June 28, 2023

On Practical Aspects of Aggregation Defenses against Data Poisoning Attacks
Wenxiao Wang, Soheil Feizi
Poisoning Attack Data Poisoning Attack Practical Application Aggregation Defense Deep Combinatorial Aggregation

March 6, 2023

Learning to Backdoor Federated Learning
Henger Li, Chen Wu, Sencun Zhu, Zizhan Zheng
LeArning Abstract Federated Learning Backdoor Attack Backdoor Learning Attack Performance Attack Framework Aggregation Defense

Aggregation Defense

Papers

Fake or Compromised? Making Sense of Malicious Clients in Federated Learning

Adversarial Robustness Unhardening via Backdoor Attacks in Federated Learning

On Practical Aspects of Aggregation Defenses against Data Poisoning Attacks

Learning to Backdoor Federated Learning