Multiple Instance Learning for Emotion Recognition Using Physiological Signals

The problem of continuous emotion recognition has been the subject several studies. proposed affective computing approaches employ sequential machine learning algorithms for improving classification stage, accounting time ambiguity emotional responses. Modeling and predicting state over is not a trivial because data labeling costly always feasible. This crucial issue in real-life applications, where sparse possibly captures only most important events rather than typical subtle changes that occur. In this work, we introduce framework from literature called Multiple Instance Learning, which able to model intervals by capturing presence or absence relevant states, without need label responses continuously (as required standard approaches). choice offers viable natural solution weakly supervised setting, taking into account We demonstrate reliability approach gold-standard scenario towards real-world usage employing an existing dataset (DEAP) purposely built one (Consumer). also outline advantages method with respect algorithms.