Game-theoretic distributed learning of generative models for heterogeneous data collections

Dmitrij Schlesinger 1 , Boris Flach 2 · 2025

Read Game-theoretic distributed learning of generative models for heterogeneous data collections by Dmitrij Schlesinger 1 , Boris Flach 2 (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-2511-01740.txt.

Description

Dmitrij Schlesinger and Boris Flach reframe distributed learning as a game. The obstacle they target is heterogeneity: different sites hold different kinds of data and run different local models, so the usual trick of averaging model parameters breaks down. Their alternative is to exchange synthetic data instead of parameters.

Each local model is a black box. It can learn from data and generate fresh data from what it learned. Semi-supervised local models even let sites work on different probability spaces. Those spaces can cover different modalities. Cast as a cooperative game, the learning problem has, for exponential-family local models, a unique Nash equilibrium, and the proposed procedure provably converges to it. Tests on standard vision benchmarks cover image classification and conditional generation, showing the approach's advantages.

Catalog record

CREATORDmitrij Schlesinger 1 , Boris Flach 2
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01740

Browse the full library · About the Tower of Babel