2024 journal article

Magnitude Matters: Fixing signSGD Through Magnitude-Aware Sparsification and Error Feedback in the Presence of Data Heterogeneity

IEEE TRANSACTIONS ON SIGNAL PROCESSING, 72, 5140–5155.

author keywords: Training; Convergence; Servers; Quantization (signal); Vectors; Training data; Signal processing algorithms; Data heterogeneity; federated learning; gradient compression; sparsification
Source: Web Of Science
Added: November 25, 2024