Models Got Talent: Identifying High Performing Wearable Human Activity Recognition Models Without Training
This work addresses the computational burden in model selection for wearable human activity recognition, offering a practical solution for real-world applications, though it is incremental as it adapts existing proxies to a new domain.
The paper tackles the problem of computationally expensive Neural Architecture Search for wearable human activity recognition by applying Zero Cost Proxies, which identify high-performing models with minimal training, achieving within 5% of full-scale training performance on six benchmark datasets.
A promising alternative to the computationally expensive Neural Architecture Search (NAS) involves the development of Zero Cost Proxies (ZCPs), which correlate well with 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.