FedDyMem: Efficient Federated Learning with Dynamic Memory and Memory-Reduce for Unsupervised Image Anomaly Detection

TL;DR

This work tackles privacy-preserving unsupervised image anomaly detection in a federated setting with heterogeneous intra-client distributions. It introduces FedDyMem, a memory-based framework where each client maintains a dynamic memory bank that is updated locally and then compressed via memory-reduce before being aggregated on a server with K-means to form a global memory shared by all clients. A memory generator and a metric loss align local memory distributions across clients, while the memory-reduce stage reduces communication overhead and privacy leakage. Experiments across six industrial and medical benchmarks show state-of-the-art AUROC performance, faster convergence, and robust privacy protection, highlighting the practical impact of memory-based knowledge sharing over parameter aggregation in federated unsupervised anomaly detection.

Abstract

Unsupervised image anomaly detection (UAD) has become a critical process in industrial and medical applications, but it faces growing challenges due to increasing concerns over data privacy. The limited class diversity inherent to one-class classification tasks, combined with distribution biases caused by variations in products across and within clients, poses significant challenges for preserving data privacy with federated UAD. Thus, this article proposes an efficient federated learning method with dynamic memory and memory-reduce for unsupervised image anomaly detection, called FedDyMem. Considering all client data belongs to a single class (i.e., normal sample) in UAD and the distribution of intra-class features demonstrates significant skewness, FedDyMem facilitates knowledge sharing between the client and server through the client's dynamic memory bank instead of model parameters. In the local clients, a memory generator and a metric loss are employed to improve the consistency of the feature distribution for normal samples, leveraging the local model to update the memory bank dynamically. For efficient communication and data privacy, a memory-reduce method based on weighted averages is proposed to significantly decrease the scale of memory banks. This reduced representation inherently, thereby mitigating the risk of data reconstruction. On the server, global memory is constructed and distributed to individual clients through k-means aggregation. Experiments conducted on six industrial and medical datasets, comprising a mixture of six products or health screening types derived from eleven public datasets, demonstrate the effectiveness of FedDyMem.

Figures (11)

• Figure 1: Illustration of federated learning in UAD. As shown on the left, for the same product anomaly detection, an intra-class distribution bias exists between local models, caused by each client possessing varying and incomplete subsets of product types. As shown in the top right, federated UAD aims to establish a single decision boundary for one-class classification within the global model. Examples of type differences in anomaly detection for different products are shown in the right down.
• Figure 2: Overview of our proposed FedDyMem framework.
• Figure 3: Illustration of the feature extraction process in FedDyMem.
• Figure 4: Illustration of Memory-reduce.
• Figure 5: Summary of the dataset distribution on five clients. The relative sizes of the pie in the chart illustrate the total data volume across individual clients. Distinct colors within each pie represent different product types.

Theorems & Definitions (8)

• Lemma 1: Correlation Quantitative Bound
• proof
• Theorem 1: Memory-Reduce Privacy
• proof
• Theorem 2: Round Loss Reduction
• proof
• Theorem 3: FedDyMem Convergence
• proof