HASA: Subnet Allocation for Compute-Constrained Model-Heterogeneous Federated Learning

In three linesHASA proposes subnet allocation for heterogeneous federated learning with resource and data constraints. The method assigns subnet widths based on heterogeneity scores computed from local data while enforcing fixed compute budgets. On next-word prediction (7 clients), HASA improves mean accuracy from 13.82% to 14.32% and strengthens worst-client performance.

Summary generated by Claude — human-verified