The Temporal Dominance of Sensations (TDS) method is a sensory evaluation technique that records time-varying changes in taste, aroma, and texture during food consumption, and is often used together with the Temporal Liking (TL) method, which records the temporal evolution of hedonic responses. Mathematical models that link sensory dynamics to liking dynamics are useful for identifying relationships between sensations and preference. Therefore, predictive models that accurately estimate TL curves from TDS curves are required. In this study, three models-reservoir computing, principal motion analysis, and the vector autoregressive model-were compared in terms of prediction accuracy. These models differ with respect to linearity versus nonlinearity and whether temporal causality is satisfied. A comparative evaluation based on the root mean square error (RMSE) of prediction revealed that the Reservoir Computing model, which is nonlinear and temporally causal, achieved the best performance, with an RMSE of 0.182 for TL curves on a 9-point scale.
MIYAGAWA et al. (Thu,) studied this question.