Robust Face Recognition via Adaptive Feature Reduction – In this paper, we propose a novel deep convolution neural network architecture based on the joint representation of the learned feature vectors, and the joint loss of the learned representation. The learned representation is learned in a fully supervised manner, and the training data consists of a sparse representation of the face vectors (based on the first two representations), and a sparse representation of the face data. During training, the data is transferred to another image space, which is then transferred to a new image. The learning method is shown to converge to a smooth state, which makes the proposed architecture a flexible and robust end-to-end architecture to achieve large-scale face recognition tasks at extremely low computational costs. Experimental results demonstrate that using a single-layer network performs as good as using the entire network combined with the network of single images.
A novel algorithm to analyze data set is proposed. The problem is to partition a data set into discrete units that are useful for inference. A novel formulation of the problem is proposed. A practical algorithm is developed to make use of the observed data and the resulting estimation using a convolutional neural network (CNN) is employed. Experimental results demonstrate that the proposed method performs favorably across different performance measures.
Recursive CNN: Bringing Attention to a Detail
Robust Face Recognition via Adaptive Feature Reduction
Learning from Negative News by Substituting Negative Images with Word2vec
Learning Deep Representations of Graphs with Missing EntriesA novel algorithm to analyze data set is proposed. The problem is to partition a data set into discrete units that are useful for inference. A novel formulation of the problem is proposed. A practical algorithm is developed to make use of the observed data and the resulting estimation using a convolutional neural network (CNN) is employed. Experimental results demonstrate that the proposed method performs favorably across different performance measures.