Assistant Professor, New Jersey Institute of Technology
1 paper at NeurIPS 2025
Local SGD converges faster under low second-order heterogeneity, and we prove it with tight bounds and supporting experiments.
