Transfer of multimodal emotion features in deep belief networks

In this paper, we investigate the effect of transfer of emotion-rich features between source and target networks on classification accuracy and training time in a multimodal setting for vision based emotion recognition. First, we propose emosource-a 6-layer Deep Belief Network (DBN), trained on popular emotion corpora for emotion classification. Second, we propose two 6-layer DBNs - emotarget and emotarget_ft and study the transfer of emotion features between source and target networks in a layer-by-layer fashion. To the best of our knowledge, this is the first research effort to study the transfer of emotion features layer-by-layer in a multimodal setting.