Sample Selection Using Multi-Task Autoencoders in Federated Learning with Non-IID Data

2026-04-30 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A novel sample selection method for federated learning (FL) with non-IID data is proposed, utilizing a multi-task autoencoder to assess sample contributions via loss and feature analysis. The approach integrates unsupervised outlier detection techniques like one-class support vector machine (OCSVM), isolation forest (IF), and adaptive loss threshold (AT) to filter noisy samples on client devices, managed by a central server. Additionally, a multi-class deep support vector data description (SVDD) loss, also centrally controlled, is introduced to refine feature-based sample selection. Validated on CIFAR10 and MNIST datasets under varying client numbers, non-IID distributions, and noise levels up to 40%, the methods demonstrate significant accuracy improvements. Loss-based selection achieved gains up to 7.02% on CIFAR10 with OCSVM and 1.83% on MNIST with AT, while federated SVDD loss further improved feature-based selection by up to 0.99% on CIFAR10 with OCSVM.

Key takeaway

For research scientists developing federated learning systems, integrating multi-task autoencoders with outlier detection methods like OCSVM or AT can significantly improve model accuracy, especially with non-IID data and high noise levels. You should consider implementing the proposed federated SVDD loss to further refine feature-based sample selection, enhancing overall model robustness and efficiency in real-world deployments.

Key insights

Multi-task autoencoders and outlier detection enhance federated learning performance by filtering noisy, redundant, or malicious samples.

Principles

• Sample contribution can be estimated via loss and feature analysis.
• Unsupervised outlier detection improves FL robustness.
• Centralized control can manage client-side sample filtering.

Method

A multi-task autoencoder estimates sample contributions. Unsupervised outlier detection (OCSVM, IF, AT) filters noisy samples. A federated SVDD loss enhances feature-based selection, all centrally managed.

In practice

• Apply OCSVM for significant CIFAR10 accuracy gains.
• Use AT for MNIST accuracy improvements.
• Integrate federated SVDD loss for feature-based refinement.

Topics

• Federated Learning
• Sample Selection
• Multi-Task Autoencoders
• Non-IID Data
• Outlier Detection

Code references

• eardic/FL_DPQS

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• VARS-FL: Validation-Aligned Client Selection for Non-IID Federated Learning in IoT Systems — Validation-aligned, history-aware client selection significantly improves federated learning convergence and stability in non-IID IoT environments.
• Open World Autoencoding Drift Detection with Novel Class Recognition in Tabular Non-stationary Data Streams — An unsupervised method detects concept drift via autoencoder reconstruction errors and novel classes using density estimation on proxy representations.
• A multi-task learning approach combining regression and classification tasks for joint feature selection — MTLComb balances mixed regression and classification tasks for unbiased joint feature selection by aligning regularization paths.
• Federated orthogonal learning for detection of liver lesions from multi-phase contrast-enhanced CT images — FedOG enhances federated learning for medical image segmentation by using orthogonal gradient decomposition to counter data heterogeneity and incomplete phases.
• The Data Points That Don’t Belong -And Why They Matter More Than You Think — Modern outlier detection leverages diverse techniques to find anomalies in complex, high-dimensional, and streaming datasets.
• Federated Nested Learning: Collaborative Training of Self-Referential Memories for Test-Time Adaptation — FedNL enables federated models to learn adaptive memory update rules, not just static weights, for robust test-time adaptation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.