Scalable Deep Learning for Pilot Performance Analysis Using Multimodal Physiological Time Series
Published in 2023 IEEE High Performance Extreme Computing Conference (HPEC), 2023
This paper presents algorithms to estimate the cognitive stress of pilots landing in difficult conditions from physiological and behavioral signals (EMG, ECG, gaze tracking, etc.). Multimodal timeseries data was synchronized using dynamic loading, enabling 140 GB of data to be loaded and preprocessed in 5 minutes. We benchmarked different model architectures for scalability and accuracy, evaluating TSFresh, an ensemble of traditional time series characteristics, and MINIROCKET, a fast featurizer using random convolutional kernels.
Recommended citation: N. Lee, P. W. Moore, L. J. Brattain. (2023). "Scalable Deep Learning for Pilot Performance Analysis Using Multimodal Physiological Time Series." 2023 IEEE High Performance Extreme Computing Conference (HPEC).
Download Paper