Models Got Talent: Identifying High Performing Wearable Human Activity Recognition Models Without Training

2025-11-08

AI Summary

To address the prohibitively high computational cost of neural architecture search (NAS) for human activity recognition (HAR) on wearable devices, this work introduces zero-cost proxies (ZCPs) to sensor time-series modeling—the first such application in this domain. We propose a training-free method that rapidly estimates HAR architecture performance using only a single batch of random input data and its forward-backward pass, yielding accurate accuracy predictions with strong noise robustness and real-world applicability. Evaluated across six standard HAR benchmark datasets, our approach selects architectures achieving 95% of the top-1 performance attained by fully training 1,500 randomly sampled architectures—reducing NAS search cost by orders of magnitude. This establishes a novel paradigm for automated, lightweight HAR model design under stringent resource constraints typical of edge wearable systems.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Hardware-aware ML
• Cognitive Modeling & Cognitive Systems: (Computational) Cognitive Architectures

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

A promising alternative to the computationally expensive Neural Architecture Search (NAS) involves the development of extit{Zero Cost Proxies (ZCPs)}, which correlate well to trained performance, but can be computed through a single forward/backward pass on a randomly sampled batch of data. In this paper, we investigate the effectiveness of ZCPs for HAR on six benchmark datasets, and demonstrate that they discover network architectures that obtain within 5% of performance attained by full scale training involving 1500 randomly sampled architectures. This results in substantial computational savings as high performing architectures can be discovered with minimal training. Our experiments not only introduce ZCPs to sensor-based HAR, but also demonstrate that they are robust to data noise, further showcasing their suitability for practical scenarios.

Problem

• Identifying high-performing wearable activity models without training
• Evaluating Zero-Cost Proxies effectiveness on six HAR benchmark datasets
• Demonstrating computational savings and robustness to data noise

Innovation

• Zero Cost Proxies estimate performance without training
• Single forward/backward pass evaluates architectures efficiently
• Proxies identify top models within 5% of full training