Multi-Label Zero-Shot Human Action Recognition via Joint Latent Embedding