Toward Understanding the Influence of Individual Clients in Federated Learning

TL;DR

This paper introduces Fed-Influence, a new metric to quantify individual client influence in federated learning, along with an efficient estimation algorithm that preserves privacy and works on various loss functions.

Contribution

It proposes a novel influence metric for clients in federated learning and an efficient, privacy-preserving estimation method applicable to convex and non-convex models.

Findings

The estimation method approximates Fed-Influence with small bias.

The approach requires minimal additional computation.

Fed-Influence can be used for model debugging.

Abstract

Federated learning allows mobile clients to jointly train a global model without sending their private data to a central server. Extensive works have studied the performance guarantee of the global model, however, it is still unclear how each individual client influences the collaborative training process. In this work, we defined a new notion, called {\em Fed-Influence}, to quantify this influence over the model parameters, and proposed an effective and efficient algorithm to estimate this metric. In particular, our design satisfies several desirable properties: (1) it requires neither retraining nor retracing, adding only linear computational overhead to clients and the server; (2) it strictly maintains the tenets of federated learning, without revealing any client's local private data; and (3) it works well on both convex and non-convex loss functions, and does not require the final model to be optimal. Empirical results on a synthetic dataset and the FEMNIST dataset demonstrate that our estimation method can approximate Fed-Influence with small bias. Further, we show an application of Fed-Influence in model debugging.