Bayesian Coreset Optimization for Personalized Federated Learning

Prateek Chanda, Shrey Modi, Ganesh Ramakrishnan · 2025

Read Bayesian Coreset Optimization for Personalized Federated Learning by Prateek Chanda, Shrey Modi, Ganesh Ramakrishnan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01800.txt.

Description

Prateek Chanda, Shrey Modi, and Ganesh Ramakrishnan slim down federated learning for personal devices. In federated learning, many clients train a shared model without surrendering raw data, but training on each client's full dataset every round is cumbersome. Their proposal lets each client forward updates computed from a small, weighted coreset, a representative sample of its data, instead of everything it holds.

The method is personalized, so each client's model can adapt to its own data. Theoretical analysis shows the average generalization error is minimax optimal up to logarithmic factors. It also gives a closed-form expression for the gap to vanilla federated learning. The gap is a function of the coreset weights and size. Experiments on benchmark datasets with several recent personalized architectures show significant gains over random sampling followed by federated training.

Catalog record

CREATORPrateek Chanda, Shrey Modi, Ganesh Ramakrishnan
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01800

Browse the full library · About the Tower of Babel