Decentralized Collaborative Learning of Personalized Models over Networks

TL;DR

This paper introduces two decentralized gossip algorithms enabling agents in a peer-to-peer network to collaboratively improve personalized models by sharing information, using label propagation and ADMM-based joint learning.

Contribution

The paper proposes novel asynchronous gossip algorithms for decentralized collaborative learning of personalized models, combining label propagation and ADMM optimization.

Findings

Algorithms effectively improve local models through peer communication.

Decentralized methods outperform isolated training in experiments.

Approaches are scalable and robust to network variations.

Abstract

We consider a set of learning agents in a collaborative peer-to-peer network, where each agent learns a personalized model according to its own learning objective. The question addressed in this paper is: how can agents improve upon their locally trained model by communicating with other agents that have similar objectives? We introduce and analyze two asynchronous gossip algorithms running in a fully decentralized manner. Our first approach, inspired from label propagation, aims to smooth pre-trained local models over the network while accounting for the confidence that each agent has in its initial model. In our second approach, agents jointly learn and propagate their model by making iterative updates based on both their local dataset and the behavior of their neighbors. To optimize this challenging objective, our decentralized algorithm is based on ADMM.