Event-Driven Online Vertical Federated Learning

2025-06-17

International Conference on Learning Representations

AI Summary

This paper addresses practical challenges in vertical federated learning (VFL), where clients possess non-overlapping features, data streams arrive asynchronously, and updates are triggered by local events—scenarios poorly supported by existing synchronous or time-driven online VFL frameworks. Method: We propose the first event-driven online VFL framework. It formally models event-asynchrony in VFL, introduces Dynamic Local Regret (DLR) as a novel performance metric, and establishes the first rigorous convergence theory for non-convex, non-stationary settings. We further design an event-triggered activation mechanism and an asynchronous collaborative optimization algorithm. Contribution/Results: We theoretically prove an upper bound on DLR that converges over time. Experiments demonstrate that our method achieves superior model stability under non-stationary data compared to state-of-the-art online VFL approaches, while significantly reducing communication overhead and computational cost.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Learning & Optimization for CV
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing

Abstract

Online learning is more adaptable to real-world scenarios in Vertical Federated Learning (VFL) compared to offline learning. However, integrating online learning into VFL presents challenges due to the unique nature of VFL, where clients possess non-intersecting feature sets for the same sample. In real-world scenarios, the clients may not receive data streaming for the disjoint features for the same entity synchronously. Instead, the data are typically generated by an emph{event} relevant to only a subset of clients. We are the first to identify these challenges in online VFL, which have been overlooked by previous research. To address these challenges, we proposed an event-driven online VFL framework. In this framework, only a subset of clients were activated during each event, while the remaining clients passively collaborated in the learning process. Furthermore, we incorporated emph{dynamic local regret (DLR)} into VFL to address the challenges posed by online learning problems with non-convex models within a non-stationary environment. We conducted a comprehensive regret analysis of our proposed framework, specifically examining the DLR under non-convex conditions with event-driven online VFL. Extensive experiments demonstrated that our proposed framework was more stable than the existing online VFL framework under non-stationary data conditions while also significantly reducing communication and computation costs.

Problem

• Address asynchronous data streaming in online VFL
• Optimize client activation via event-driven collaboration
• Handle non-convex models in non-stationary environments

Innovation

• Event-driven online VFL framework
• Dynamic local regret integration
• Subset client activation per event