Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions

Collection of recent ML/stat arXiv papers (May 29, 2026) covering: (1) dynamics of stochastic momentum under sparse updates in high dimensions; (2) an anytime-valid extension of Federated Conformal RAG (Anytime-FC-RAG) for LLM swarms with sequential guarantees and bandwidth savings; (3) multi-task prediction-powered inference for improved statistical power with scarce labels; (4) deep individualized treatment-rule learning for bivariate survival outcomes; (5) tight rates and optimal per-node bandwidth allocation for federated probe-logit distillation (FPLD) under heterogeneous budgets; (6) RMT