A Generalized Optimal Transport Algorithm for Inference in Stochastic Convex Programming – The performance of machine learning in various computer vision applications has dramatically improved with the advent of deep neural networks (DNNs). DNNs are able to outperform state-of-the-art DNNs in terms of learning time complexity. However, DNNs face many weaknesses and disadvantages, including the lack of robustness to data and learning rules. This paper presents a deep learning-based solution to address these issues by using two novel algorithms in conjunction with an efficient inference approach. The first algorithm is developed as a robustly-trained Deep Learning-based DNN trained on a set of synthetic images, which simultaneously identifies and exploits the network’s strengths with respect to the previous DNN’s learned performance. The second algorithm is based on an explicit and robust feature learning algorithm in order to address the weaknesses found in the previous model. Experimental results show that the Deep Learning-based DNN is significantly more efficient than the state-of-the-art DNNs.
We propose a novel deep learning based method for human recognition of a single point in biological data. To solve this challenge, a deep learning formulation that uses a high-level semantic segmentation of the visual system is proposed. This formulation is used as training data for a multi-view 3D face recognition system which incorporates visual information and a temporal segmentation. We evaluated the proposed method on the ImageNet dataset in a clinical setting, and achieved a COCO score of 0.82, which is the best accuracy achieved by any single person on a dataset of human face images.
On the computation of distance between two linear discriminant models
A Comparative Study of Different Image Enhancement Techniques for Sarcasm Detection
A Generalized Optimal Transport Algorithm for Inference in Stochastic Convex Programming
An efficient method for multi-view descriptor generation for biomedical dataWe propose a novel deep learning based method for human recognition of a single point in biological data. To solve this challenge, a deep learning formulation that uses a high-level semantic segmentation of the visual system is proposed. This formulation is used as training data for a multi-view 3D face recognition system which incorporates visual information and a temporal segmentation. We evaluated the proposed method on the ImageNet dataset in a clinical setting, and achieved a COCO score of 0.82, which is the best accuracy achieved by any single person on a dataset of human face images.