Communication Compression for Byzantine Robust Learning: New Algorithms and Rates
The authors introduce new Byzantine-robust methods with communication compression, showcasing improved convergence rates and tolerance to Byzantine workers under over-parametrization.