ProbSelect: Stochastic Client Selection for GPU-Accelerated Compute Devices in the 3D Continuum

2025-11-11

AI Summary

In the dynamic edge-cloud-air continuum—particularly involving satellites and mobile devices—client selection for federated learning faces challenges in simultaneously ensuring real-time responsiveness, accommodating GPU heterogeneity (e.g., NVIDIA vs. AMD), and meeting stringent service-level objective (SLO) guarantees. This paper proposes a lightweight, history- and monitoring-free client selection framework. It is the first to jointly integrate analytical modeling and probabilistic prediction to explicitly characterize the latency-energy-accuracy trade-offs and their uncertainties under GPU-accelerated training. The framework enables cross-architecture, multi-workload adaptive selection under user-specified SLO constraints. Experiments demonstrate that, compared to baseline methods, our approach improves SLO compliance by 13.77% on average and reduces computational resource waste by 72.5%.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• 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
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Integration of edge, cloud and space devices into a unified 3D continuum imposes significant challenges for client selection in federated learning systems. Traditional approaches rely on continuous monitoring and historical data collection, which becomes impractical in dynamic environments where satellites and mobile devices frequently change operational conditions. Furthermore, existing solutions primarily consider CPU-based computation, failing to capture complex characteristics of GPU-accelerated training that is prevalent across the 3D continuum. This paper introduces ProbSelect, a novel approach utilizing analytical modeling and probabilistic forecasting for client selection on GPU-accelerated devices, without requiring historical data or continuous monitoring. We model client selection within user-defined SLOs. Extensive evaluation across diverse GPU architectures and workloads demonstrates that ProbSelect improves SLO compliance by 13.77% on average while achieving 72.5% computational waste reduction compared to baseline approaches.

Problem

• Addresses client selection challenges in federated learning across edge-cloud-space continuum
• Overcomes impracticality of continuous monitoring in dynamic satellite and mobile environments
• Optimizes GPU-accelerated training selection without historical data requirements

Innovation

• ProbSelect uses analytical modeling for client selection
• It employs probabilistic forecasting without historical data
• Focuses on GPU-accelerated devices across 3D continuum