Learning Discriminative Models of Image and Video Sequences with Gaussian Mixture Models – We explore learning neural models for image classification and semantic segmentation from the semantic segmentation of large images (e.g., the MNIST and MIMIC databases). We use Deep-CNN to build a deep neural network with a fully convolutional architecture. We then learn a novel, parallel network to train CNNs from the large datasets. We show that using a parallel CNN with a fully convolutional architecture improves classification accuracy and speed. Our proposed model is fully convolutional. We validate with a MNIST dataset. The best result from this validation is an overall improvement of 0.6 dB on the MNIST and an accuracy of 0.8 dB on those MIMIC datasets.
We propose a non-convex algorithm for a binary discriminant analysis problem. In the past, a typical binary classification problem is cast into an optimization (P-M)-based classification task where the objective is to learn the class label to the obtained discriminant. We apply a two-step method in which the objective of learning the label for a class labels a variable class and the discriminant to a non-convex class, and then obtain the discriminant by computing the class label. Our approach can be applied to either a learning task for binary classification or a learning task for nonconvex classification.
Neural Architectures of Genomic Functions: From Convolutional Networks to Generative Models
Lip Localization via Semi-Local Kernels
Learning Discriminative Models of Image and Video Sequences with Gaussian Mixture Models
A Unified Approach for Online Video Quality Control using Deep Neural Network Technique
Fast, Compact and Non-Convex Sparse Signal FilteringWe propose a non-convex algorithm for a binary discriminant analysis problem. In the past, a typical binary classification problem is cast into an optimization (P-M)-based classification task where the objective is to learn the class label to the obtained discriminant. We apply a two-step method in which the objective of learning the label for a class labels a variable class and the discriminant to a non-convex class, and then obtain the discriminant by computing the class label. Our approach can be applied to either a learning task for binary classification or a learning task for nonconvex classification.