Decentralized SGD under Heavy-Tailed Noise: Optimal Convergence Rates and the Role of Gradient Clipping
Clipped decentralized SGD reaches order-optimal rates under heavy-tailed noise with linear agent speed-up.
Stories per day · 30 days
3 stories, newest first · one card per story
Stories tagged #convergence
AI is exposing a security structure built for yesterday’s threats
EY's Jeffrey Sallet argues AI-driven deepfakes and impersonation require integrating cybersecurity, physical security, HR and legal functions.
1
Quantile-based Loss Filtering for Outlier-Robust Stochastic Gradient Descent
Quantile-k-Loss SGD filters corrupted component losses by quantile sampling, proving linear convergence while outperforming standard and min-k-loss SGD.