Emotion-Aroused Human Behaviors Perception Using RNNPB

This paper proposes a novel framework to recognize emotions using the algorithm of a recurrent neural network with parameter bias (RNNPB). For this purpose, we performed three simulation experiments aim to explore the relationship between the perception and action. Three types of emotion-driven sequences are fed into the network for training. The training of RNNPB utilizes back-propagation through time (BPTT) method and the parametric bias unit (PB unit) updates in a self-organizing way. The results of the experiments show that the merged sequences can distinguish the emotion better compared to the other two kinds of information.