Federated Deconfounding and Debiasing Learning for Out-of-Distribution Generalization

2025-05-08

AI Summary

In federated learning (FL), attribute bias induces spurious, non-causal associations, leading to inconsistent local model optimization and failure in out-of-distribution generalization. To address this, we propose the first structured causal graph modeling framework tailored for FL, jointly enabling intra-client deconfounding—via backdoor adjustment and counterfactual sample generation—and inter-client debiased learning—through causal prototype regularization that bridges heterogeneous representations and suppresses background confounders. Our approach innovatively integrates causal inference with federated knowledge distillation, establishing an interpretable and robust invariant representation learning paradigm. Evaluated on two benchmark datasets, our method achieves an average Top-1 accuracy gain of 4.5% over nine state-of-the-art baselines, significantly enhancing model focus on target objects under unseen distributions and improving generalization robustness.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Adversarial Attacks & Robustness
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Attribute bias in federated learning (FL) typically leads local models to optimize inconsistently due to the learning of non-causal associations, resulting degraded performance. Existing methods either use data augmentation for increasing sample diversity or knowledge distillation for learning invariant representations to address this problem. However, they lack a comprehensive analysis of the inference paths, and the interference from confounding factors limits their performance. To address these limitations, we propose the underline{Fed}erated underline{D}econfounding and underline{D}ebiasing underline{L}earning (FedDDL) method. It constructs a structured causal graph to analyze the model inference process, and performs backdoor adjustment to eliminate confounding paths. Specifically, we design an intra-client deconfounding learning module for computer vision tasks to decouple background and objects, generating counterfactual samples that establish a connection between the background and any label, which stops the model from using the background to infer the label. Moreover, we design an inter-client debiasing learning module to construct causal prototypes to reduce the proportion of the background in prototype components. Notably, it bridges the gap between heterogeneous representations via causal prototypical regularization. Extensive experiments on 2 benchmarking datasets demonstrate that methodname{} significantly enhances the model capability to focus on main objects in unseen data, leading to 4.5% higher Top-1 Accuracy on average over 9 state-of-the-art existing methods.

Problem

• Addresses attribute bias in federated learning causing inconsistent local model optimization
• Eliminates confounding paths via structured causal graph and backdoor adjustment
• Enhances model focus on main objects in unseen data for better accuracy

Innovation

• Constructs causal graph for inference analysis
• Performs backdoor adjustment to eliminate confounders
• Uses causal prototypes for debiasing regularization