Learned Neighbor Trust for Collaborative Deployment in Model-Agnostic Decentralized Learning

TL;DR

This paper introduces LNTrust, a decentralized learning method where nodes learn to trust neighbors, improving collaborative deployment accuracy with less communication in heterogeneous IoT settings.

Contribution

It proposes a novel trust function learning approach that enhances model collaboration during training and deployment in decentralized, model-agnostic environments.

Findings

LNTrust outperforms output-only baselines in accuracy across multiple datasets.

It reduces communication overhead compared to previous decentralized methods.

The trust mechanism effectively adapts to heterogeneous and resource-constrained settings.

Abstract

Many decentralized distillation methods are designed around training-time coordination, yet deploy each node in isolation even when more capable neighbors remain available at inference time. This is an incomplete objective for settings such as IoT, where devices are heterogeneous, data is scarce and skewed, and a node's strongest neighbors may far exceed its own local capacity. We study how nodes should train so that their predictions compose well at deployment, and how each node should learn whom to trust. Under a server-free, model-agnostic protocol where nodes exchange only queries and soft predictions, we propose Learned Neighbor Trust (LNTrust) wherein each node learns a compact trust function over its neighborhood from local validation evidence. This trust function gates auxiliary distillation during training and defines a deployment ensemble at inference, so that collaboration learned during training transfers directly to deployment. Across datasets and topologies, LNTrust improves deployed accuracy over the strongest output-only baseline by large margins while using significantly less communication than previous methods.