Learn How to Query from Unlabeled Data Streams in Federated Learning
This addresses the problem of costly data annotation in federated learning for decentralized clients, offering an incremental improvement by optimizing sample selection locally to enhance global training.
The paper tackles the challenge of selecting informative unlabeled data for annotation in federated learning, where data arrives as streams without labels, and proposes LeaDQ, a method using multi-agent reinforcement learning to improve global model accuracy, achieving performance gains over benchmarks in image and text tasks.
Federated learning (FL) enables collaborative learning among decentralized clients while safeguarding the privacy of their local data. Existing studies on FL typically assume offline labeled data available at each client when the training starts. Nevertheless, the training data in practice often arrive at clients in a streaming fashion without ground-truth labels. Given the expensive annotation cost, it is critical to identify a subset of informative samples for labeling on clients. However, selecting samples locally while accommodating the global training objective presents a challenge unique to FL. In this work, we tackle this conundrum by framing the data querying process in FL as a collaborative decentralized decision-making problem and proposing an effective solution named LeaDQ, which leverages multi-agent reinforcement learning algorithms. In particular, under the implicit guidance from global information, LeaDQ effectively learns the local policies for distributed clients and steers them towards selecting samples that can enhance the global model's accuracy. Extensive simulations on image and text tasks show that LeaDQ advances the model performance in various FL scenarios, outperforming the benchmarking algorithms.