Spatially Aware Convolutional Neural Networks for Person Re-Identification – This paper presents a novel method for supervised learning for face detection. The method first learns a similarity graph from labeled face images, or RGB images. Then, we learn a similarity graph for segmentation that is based on a novel feature vector representation. After the segmentation, training for a face detection problem is formulated in terms of the discriminative similarity of images from different classes. We achieve this by leveraging recent advances in deep learning as well as the recently proposed Neural Network-Aided Perceptron (NNAP) method. This method works on both visual and physiological datasets. We show how the network can be used to successfully perform face detection in two scenarios: visual face detection and an adaptive face tracker. Our preliminary method achieves state-of-the-art accuracies of ~83% on the MNIST and ~97% on the TIMIT dataset, and shows promising results on both visual and physiological datasets.
In this paper, we propose a novel method for performing clustering of graph-structured data from multiple data sources, using several approaches including the use of clustering, deep learning, Bayesian networks, and conditional random fields. Furthermore, we provide a numerical example where a standard Bayesian network is used. The proposed method is very simple and has theoretical support beyond the classical methods, i.e., the proposed method performs clustering without any supervision. The implementation of the method is carried out using a large-scale dataset and has been extensively evaluated on two publicly available datasets. The experimental results on these datasets clearly indicate the usefulness of the proposed method to improve the performance of graph-structured data.
The Complexity of Logics in Redistributing Knowledge
Graphical Models Under Uncertainty
Spatially Aware Convolutional Neural Networks for Person Re-Identification
Efficient Online Convex Optimization with a Non-Convex Cost Function
Clustering on multiple graph connectionsIn this paper, we propose a novel method for performing clustering of graph-structured data from multiple data sources, using several approaches including the use of clustering, deep learning, Bayesian networks, and conditional random fields. Furthermore, we provide a numerical example where a standard Bayesian network is used. The proposed method is very simple and has theoretical support beyond the classical methods, i.e., the proposed method performs clustering without any supervision. The implementation of the method is carried out using a large-scale dataset and has been extensively evaluated on two publicly available datasets. The experimental results on these datasets clearly indicate the usefulness of the proposed method to improve the performance of graph-structured data.