Pain assessment improved with faster neural design sharing method
Efficient Architecture Search under Leave-One-Subject-Out Evaluation
Machine Learning
Summary
Automated systems can detect pain by analyzing signals with deep neural networks, but designing these networks usually takes a lot of time and effort. The authors found a way to speed up this design process by sharing parts of the search across different people, cutting the work needed when testing on many subjects. Their method uses fewer searches and still works just as well in identifying pain signals. This makes building pain detection systems more efficient without losing accuracy.
What this means in practice
- •For medical device developers: Design efficient pain detection models that perform well across different patients with less computational effort.
- •For wearable technology engineers: Develop wearable sensors with accurate pain monitoring that require fewer and lighter models, extending device battery life and usability.
Tested on one dataset.
Authors
Heinke Hihn, Friedhelm Schwenker
Abstract
Deep neural architectures are widely used for signal processing in automated pain assessment systems. However, architecture design has remained largely a manual task despite the potential efficiency benefits of Neural Architecture Search (NAS). Embedding NAS in a Leave-One-Subject-Out (LOSO) evaluation is computationally demanding because a fully nested implementation requires $N$ independent architecture searches and, assuming approximately linear training cost, scales as $\mathcal{O}(N^2)$. We propose a block-based, leakage-controlled approach that shares NAS runs between subjects, reducing the number of searches from $N$ to $B$, where $B \ll N$, dubbed PainNAS. On the BioVid Heat Pain dataset, PainNAS yields comparable subject-level accuracy with substantially fewer parameters and FLOPs.