Byzantine Failure

Byzantine failures, where malicious or faulty nodes in a distributed system provide incorrect information, pose a significant challenge to the reliability of distributed machine learning. Current research focuses on developing robust algorithms and aggregation methods, such as those employing trimmed means, medians, or momentum filtering, to mitigate the impact of these failures in various settings, including federated learning and finite mixture models. These advancements are crucial for ensuring the accuracy and security of large-scale machine learning systems, particularly in scenarios with unreliable or adversarial participants. The development of Byzantine-robust algorithms is driving improvements in the scalability and trustworthiness of distributed computation across diverse applications.

Papers

October 21, 2024

IBGP: Imperfect Byzantine Generals Problem for Zero-Shot Robustness in Communicative Multi-Agent Systems
Yihuan Mao, Yipeng Kang, Peilun Li, Ning Zhang, Wei Xu, Chongjie Zhang
Software Defined Consensus Mechanism Byzantine Fault Tolerance Zero Shot Adversarial Robustness Byzantine Failure

September 13, 2024

Byzantine-Robust and Communication-Efficient Distributed Learning via Compressed Momentum Filtering
Changxin Liu, Yanghao Li, Yuhao Yi, Karl H. Johansson
LeArning Abstract Communication Efficient Byzantine Robust Momentum Based Byzantine Failure

July 19, 2024

Byzantine-tolerant distributed learning of finite mixture models
Qiong Zhang, Jiahua Chen
LeArning Abstract Split Learning Byzantine Robust Finite Mixture Byzantine Failure

April 9, 2024

Aggressive or Imperceptible, or Both: Network Pruning Assisted Hybrid Byzantines in Federated Learning
Emre Ozfatura, Kerem Ozfatura, Alptekin Kupcu, Deniz Gunduz
Network Pruning Byzantine Attack Imperceptible Pattern Imperceptible Attack Byzantine Failure

February 5, 2024

Dynamic Byzantine-Robust Learning: Adapting to Switching Byzantine Workers
Ron Dorfman, Naseem Yehya, Kfir Y. Levy
Learning Rate Byzantine Robust Byzantine Seal Byzantine Failure Time Varying Impedance Attack

July 25, 2023

High Dimensional Distributed Gradient Descent with Arbitrary Number of Byzantine Attackers
Puning Zhao, Zhiguo Wan
Gradient Descent Structured Output High Dimension Mean Estimation Byzantine Attack High Dimensional Problem Byzantine Failure Gradient Vector

June 1, 2023

Byzantine-Robust Clustered Federated Learning
Zhixu Tao, Kun Yang, Sanjeev R. Kulkarni
Adversarial Attack Byzantine Robust Federated Learning Byzantine Failure

April 15, 2023

Practical Differentially Private and Byzantine-resilient Federated Learning
Zihang Xiang, Tianhao Wang, Wanyu Lin, Di Wang
Practical Application Private Stochastic Byzantine Robust Federated Learning Byzantine Robust Aggregation Byzantine Resilience Byzantine Failure

February 12, 2023

Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization
Hamidreza Almasi, Harsh Mishra, Balajee Vamanan, Sathya N. Ravi
Training Data Potential Scalability Convex Optimization Convergence Guarantee Subspace Learning Country Recognition Planning Loss Byzantine Robust Aggregation Byzantine Failure One Shot Aggregation

November 24, 2021

One More Step Towards Reality: Cooperative Bandits with Imperfect Communication
Udari Madhushani, Abhimanyu Dubey, Naomi Ehrich Leonard, Alex Pentland
Real World Minimax Optimization Limited Communication Cooperative Bandit Byzantine Failure

Byzantine Failure

Papers

IBGP: Imperfect Byzantine Generals Problem for Zero-Shot Robustness in Communicative Multi-Agent Systems

Byzantine-Robust and Communication-Efficient Distributed Learning via Compressed Momentum Filtering

Byzantine-tolerant distributed learning of finite mixture models

Aggressive or Imperceptible, or Both: Network Pruning Assisted Hybrid Byzantines in Federated Learning

Dynamic Byzantine-Robust Learning: Adapting to Switching Byzantine Workers

High Dimensional Distributed Gradient Descent with Arbitrary Number of Byzantine Attackers

Byzantine-Robust Clustered Federated Learning

Practical Differentially Private and Byzantine-resilient Federated Learning

Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization

One More Step Towards Reality: Cooperative Bandits with Imperfect Communication