#artificial intelligence
Dec 2024
dSTAR: Straggler Tolerant and Byzantine Resilient Distributed SGD
DSTAR is proposed, a lightweight and efficient approach for distributed stochastic gradient descent that enhances robustness and convergence and outperforms other Byzantine-resilient methods that often suffer up to a 40-50\% accuracy drop under attack.
Jia Yan, Pratik Chaudhari, Leonard Kleinrock
· arXiv.org · 0 citations