Contrastive Self-Supervised Learning at the Edge: An Energy Perspective

2025-10-09

AI Summary

Deploying contrastive self-supervised learning (SSL) on edge/fog devices faces challenges of high energy consumption, memory constraints, and limited labeled data. Method: This paper introduces the first systematic benchmarking framework tailored for edge scenarios, integrating fine-grained energy profiling and few-shot training evaluation. Experiments evaluate four mainstream contrastive SSL methods—SimCLR, MoCo, SimSiam, and Barlow Twins—using lightweight backbone networks across varying data scales on real edge hardware. Contribution/Results: SimCLR achieves the lowest energy consumption while maintaining competitive representation quality, significantly outperforming alternatives. Its low computational overhead and strong robustness under few-shot conditions make it the most suitable candidate for edge deployment. The study establishes a reproducible evaluation paradigm and delivers key design insights for practical contrastive SSL in resource-constrained environments.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Adversarial Attacks & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Cloud, edge and content delivery systems for the Web
• Security and Privacy: Large-scale security measurements

Abstract

While contrastive learning (CL) shows considerable promise in self-supervised representation learning, its deployment on resource-constrained devices remains largely underexplored. The substantial computational demands required for training conventional CL frameworks pose a set of challenges, particularly in terms of energy consumption, data availability, and memory usage. We conduct an evaluation of four widely used CL frameworks: SimCLR, MoCo, SimSiam, and Barlow Twins. We focus on the practical feasibility of these CL frameworks for edge and fog deployment, and introduce a systematic benchmarking strategy that includes energy profiling and reduced training data conditions. Our findings reveal that SimCLR, contrary to its perceived computational cost, demonstrates the lowest energy consumption across various data regimes. Finally, we also extend our analysis by evaluating lightweight neural architectures when paired with CL frameworks. Our study aims to provide insights into the resource implications of deploying CL in edge/fog environments with limited processing capabilities and opens several research directions for its future optimization.

Problem

• Evaluating energy efficiency of contrastive learning frameworks for edge deployment
• Addressing computational challenges of self-supervised learning on resource-constrained devices
• Benchmarking memory usage and data requirements for edge-based contrastive learning

Innovation

• Benchmarking four contrastive learning frameworks for edge deployment
• SimCLR achieves lowest energy consumption in various data conditions
• Evaluating lightweight neural architectures with contrastive learning frameworks