Scalable Algorithms for Learning Low-rank Mixtures with Large-Margin Classification – This paper presents a methodology for a hierarchical clustering model for classification tasks that use two or more classes. The class-specific clustering model is based in hierarchical clustering and can also be used to predict the clustering probability. The model can be used for all scenarios in which it is more suitable as a tool for clustering data.

Many current approaches to multi-scale data collection have been limited to a set of random points that are spatially aligned and, in general, they are too sensitive to local minima between the measurements. In this paper, we develop a novel method to extract spatially-aligned points from the same measurements for multiple datasets. The main contribution is that our method allows for an easily scalable approach to combining multiple measurements to extract spatially-aligned points. A simple yet accurate algorithm, which can handle sparse measurements and sparse minima, was validated on the MNIST dataset and compared against other approaches. In particular, the proposed method outperforms the others in both test-time estimation and classification on the MNIST dataset and compared to other methods.

On the Role of Context Dependency in the Performance of Convolutional Neural Networks in Task 1 Problems

Visual Tracking using Visual Tensor Factorization with Applications to Automated Vehicle Analysis and Tracking

Scalable Algorithms for Learning Low-rank Mixtures with Large-Margin Classification

Semantic Regularities in Textual-Visual Embedding

A Structural-Time Decomposition of the Distance Between a 3D Surface and the Grade BarMany current approaches to multi-scale data collection have been limited to a set of random points that are spatially aligned and, in general, they are too sensitive to local minima between the measurements. In this paper, we develop a novel method to extract spatially-aligned points from the same measurements for multiple datasets. The main contribution is that our method allows for an easily scalable approach to combining multiple measurements to extract spatially-aligned points. A simple yet accurate algorithm, which can handle sparse measurements and sparse minima, was validated on the MNIST dataset and compared against other approaches. In particular, the proposed method outperforms the others in both test-time estimation and classification on the MNIST dataset and compared to other methods.