[4K3-GS-1] Fundamental AI, theory: cognition / perception

[4K3-GS-1-01] Estimating individual differences in movie preference via brain-activity prediction using convolutional neural networks

〇Kiichi Kawahata^1,2, Antoine Blanc², Shinji Nishimoto^1,2, Satoshi Nishida^1,2 (1. Graduate School of Frontier Biosciences, Osaka University, 2. Center for Information and Neural Networks (CiNet), Advanced ICT Research Institute, National Institute of Information and Communications Technology (NICT))

Keywords:Deep Learning, Brain, Preference, Individual Differences, Neuroimaging

Recently, machine-learning techniques have been developed to estimate individual differences in subjective preferences for movies. This study attempted to improve the performance of convolutional neural networks (CNNs) in estimating preferences for movies by incorporating individual brain information into the CNNs. To this end, we introduced a method to estimate subjective preferences for movies from brain activity predicted using movie-evoked activation patterns in CNN hidden layers. This prediction process corresponds to the transformation from CNN features to brain feature representations. We compared the proposed method with the direct estimation from CNNs in terms of their performance in subjective-preference estimation. As a result, the performance of the proposed method was significantly higher than that of the CNN direct estimation. This result suggests that the performance of machine-learning techniques in estimating subjective preferences improves by incorporating the representational characters of individual brains into the techniques.