Security of Distributed Gradient Descent Against Byzantine Agents
This paper investigates the security of Decentralized Gradient Descent (DGD) algorithms on undirected graphs. Specifically, we consider Byzantine agents that inject stealthy attacks to degrade the performance of the DGD algorithm in terms of its optimal value. To mitigate the effect of stealthy attacks, some agents are allocated to monitor the evolution of their gradient. We then propose a method to quantify the maximum deviation caused by the Byzantine agent in the optimal value of the DGD. Our approach serves as a security metric to evaluate the robustness of graph structures against Byzantine attacks. We validate our findings through numerical simulations.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Byzantine-Robust Distributed Learning: Towards Optimal Statistical Rates
In large-scale distributed learning, security issues have become increasingly important. Particularly in a decentralized environment, some computing units may behave abnormally, or even exhibit Byzantine failures -- arbi…
Byzantine Fault-Tolerant Distributed Machine Learning Using Stochastic Gradient Descent (SGD) and Norm-Based Comparative Gradient Elimination (CGE)
This paper considers the Byzantine fault-tolerance problem in distributed stochastic gradient descent (D-SGD) method - a popular algorithm for distributed multi-agent machine learning. In this problem, each agent samples…
ByGARS: Byzantine SGD with Arbitrary Number of Attackers
We propose two novel stochastic gradient descent algorithms, ByGARS and ByGARS++, for distributed machine learning in the presence of any number of Byzantine adversaries. In these algorithms, reputation scores of workers…
Communication-Efficient and Byzantine-Robust Distributed Learning with Error Feedback
We develop a communication-efficient distributed learning algorithm that is robust against Byzantine worker machines. We propose and analyze a distributed gradient-descent algorithm that performs a simple thresholding ba…
Utilizing Redundancy in Cost Functions for Resilience in Distributed Optimization and Learning
This paper considers the problem of resilient distributed optimization and stochastic machine learning in a server-based architecture. The system comprises a server and multiple agents, where each agent has a local cost …
Distributed Optimization