Adaptive Data Admission and Retention for Streaming Federated Learning

2026-07-27

AI Summary

This work addresses streaming federated learning under stringent client-side memory and sampling cost constraints by proposing a joint framework for server-side data admission and client-side memory management. The framework dynamically selects and retains incoming data to minimize cumulative excess population risk. Its core innovation lies in the first decomposition of effective sample size into three components: instantaneous scale, independent growth, and reuse imbalance. Guided by this decomposition, the authors design an active constraint-drift strategy based on a surrogate penalty function, a K-step retention rule, and a time-varying rectangular admission region. Theoretical analysis establishes a sublinear regret bound with controllable constraint violation, while experiments demonstrate that the method closely approaches the performance of an unconstrained ideal benchmark while strictly adhering to resource limitations.

Abstract

We study streaming federated learning with limited client memory, where newly generated training data incur time-varying sampling costs and must be selectively admitted and retained over time. We consider a joint server-side admission and client-side memory-management framework with the objective of minimizing the cumulative excess population risk under a sampling-cost budget and buffer constraints. We first derive a learning-error bound that explicitly captures the effects of instantaneous training sample size, distinct-sample growth, and reuse imbalance through a characterization of the effective sample size. Through a surrogate penalty obtained from this bound, we develop an Active-Constraint Drift-Plus-Penalty (ACDPP) policy that combines a structured client-side $K$-step retention rule with a server-side online admission rule and a time-varying rectangular admission region. We further present a sequence of comparison arguments, via an auxiliary constant-admission policy, that connects the ACDPP learning bound to a costless oracle benchmark. This yields explicit guarantees in terms of sublinear regret and sampling-cost violation, while the buffer-occupancy violation is controlled through offline selection of the retention horizon. Experiments on multiple datasets demonstrate that the proposed policy remains close to the oracle benchmark while satisfying the sampling-cost and buffer constraints.

Problem

• streaming federated learning
• data admission
• memory management
• sampling cost
• buffer constraint

Innovation

• Streaming Federated Learning
• Adaptive Data Admission
• Memory-Constrained Clients
• Effective Sample Size
• Drift-Plus-Penalty