Compositional Models for Video Event Detection: A Multiple Kernel Learning Latent Variable Approach | IEEE Conference Publication | IEEE Xplore