Randomized neural networks for preference learning with physiological data

© 2018 Elsevier B.V. The paper discusses the use of randomized neural networks to learn a complete ordering between samples of heart-rate variability data by relying solely on partial and subject-dependent information concerning pairwise relations between samples. We confront two approaches, i.e. Extreme Learning Machines and Echo State Networks, assessing the effectiveness in exploiting hand-engineered heart-rate variability features versus using raw beat-to-beat sequential data. Additionally, we introduce a weight sharing architecture and a preference learning error function whose performance is compared with a standard architecture realizing pairwise ranking as a binary-classification task. The models are evaluated on real-world data from a mobile application realizing a guided breathing exercise, using a dataset of over 54 K exercising sessions. Results show how a randomized neural model processing information in its raw sequential form can outperform its vectorial counterpart, increasing accuracy in predicting the correct sample ordering by about 20%. Further, the experiments highlight the importance of using weight sharing architectures to learn smooth and generalizable complete orders induced by the preference relation.