Prediction-space knowledge markets for communication-efficient federated learning on multimedia tasks

Du Wenzhang · 2025

Read Prediction-space knowledge markets for communication-efficient federated learning on multimedia tasks by Du Wenzhang (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-2512-00841.txt.

Description

Du Wenzhang proposes a market-based twist on federated learning in a 2025 paper. Federated learning trains one model across many devices or sites without pooling their data. For multimedia tasks, the method strains under uneven data across clients and tight communication limits, especially with large models.

Classic averaging methods send full model weights back and forth, and they can diverge when client data differ a lot. The proposed system, KTA v2, trades knowledge in prediction space instead of raw weights. Clients buy and sell predictions in a knowledge market. Trading compact predictions could cut traffic while keeping learning stable.

Catalog record

CREATORDu Wenzhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-00841

Browse the full library · About the Tower of Babel