Learning Multimodal Representations for Sample-efficient Recognition of Human Actions | IEEE Conference Publication | IEEE Xplore