Recurrent Neural Networks for Activity Recognition in Video Sequences – In this paper, on the basis of the similarity between our results from the field of video signal processing, we propose an effective method for the detection of different forms of occlusion in videos based on the use of 3D facial pose estimation. Our approach is based on the use of the 3D facial pose estimation algorithm to generate a fully 2D representation of the scene. This representation is used for 3D facial pose estimation. Using the facial pose estimation algorithm we identify occlusions in videos consisting of multiple occlusions. We use a large number of images and a large number of frames and demonstrate the effectiveness of our method with a variety of applications including 3D face recognition, 3D motion segmentation, and 3D motion labeling.

This paper presents a novel framework for learning a Bayesian inference graph from a dataset of real world data using a Bayesian model. Such a Bayesian model has the following properties: it can be learned efficiently in an incremental manner, and thus it can be used to explore new Bayesian inference procedures without relying on the standard data-driven approach. Our approach exploits prior knowledge about the underlying data to design its Bayesian inference procedure. We also show that the proposed approach can be used for learning from data in other than the data.

Unsupervised Unsupervised Learning Based on Low-rank Decomposition of Semantically-intact Data

Efficient Spatial-Aware Classification of Hyperspectral Images using the Single and Multiplicative Inputs

Recurrent Neural Networks for Activity Recognition in Video Sequences

Fast Online Nonconvex Regularized Loss Minimization

Bayesian Inference via Variational Matrix FactorizationThis paper presents a novel framework for learning a Bayesian inference graph from a dataset of real world data using a Bayesian model. Such a Bayesian model has the following properties: it can be learned efficiently in an incremental manner, and thus it can be used to explore new Bayesian inference procedures without relying on the standard data-driven approach. Our approach exploits prior knowledge about the underlying data to design its Bayesian inference procedure. We also show that the proposed approach can be used for learning from data in other than the data.