Expected Gain-based Escalation in Vertical Federated Learning

2026-06-30

AI Summary

This work addresses the high communication and computational overhead incurred by full-sample collaborative fusion in vertical federated learning. To mitigate this, the authors propose a two-round selective upgrade inference protocol that triggers costly embedding fusion only when the expected fusion is likely to improve prediction correctness. The method introduces an analytical routing scoring mechanism based on expected gain, eliminating the need for training an additional routing network. By integrating posterior calibration with class-level model reliability estimation, it enables interpretable and efficient sample-wise upgrade decisions. Experiments demonstrate that, under multi-view classification and view-degradation settings, the proposed approach significantly outperforms baselines relying on confidence scores, learning gains, or deferred decisions, achieving a superior trade-off between communication cost and model accuracy.

Abstract

Collaborative inference can improve predictive performance by integrating complementary information across agents, but applying collaborative fusion to every sample can incur unnecessary communication and computational overhead. This trade-off is particularly relevant in vertical federated learning (VFL), where clients observe different views of the same sample and fusion typically requires transmitting intermediate representations to a server. We study selective escalation in a two-round VFL inference protocol, in which a low-cost first round produces a prediction from client posteriors and a second embedding-fusion round is invoked only when it is expected to improve the final decision. We formulate routing as expected-gain score estimation: a sample is escalated when a predicted improvement in correctness justifies the additional communication. The proposed analytical score combines a calibrated pooled posterior with classwise reliability estimates of the VFL model, both obtained from held-out calibration data, yielding an interpretable router that requires no separately trained routing network. Experiments on multi-view classification benchmarks, including controlled test--time view degradation settings, show that the proposed router improves the communication-accuracy trade-off over confidence-, learned-gain-, and deferral-based baselines.

Problem

• vertical federated learning
• collaborative inference
• communication overhead
• selective escalation
• inference efficiency

Innovation

• vertical federated learning
• selective escalation
• expected gain
• calibrated posterior
• communication-accuracy trade-off