CORTEXA
← Browse

Radhakrishna Achanta

2 papers indexed

arxivcs.LGcs.AI2026-07-20

Federated Lightweight Fine-Tuning

Radhakrishna Achanta, Will Reed

Federated fine-tuning is bottlenecked by communication: FedAvg and pseudo-gradient schemes transmit a payload that scales with the model, and gradient compression shrinks it by only a constant factor. We take a different lever. Mapping networks generate a network's weights from a…

View free PDFSource page