Multi-dimensional Recurrent Neural Networks for Music Genome Analysis – We present a deep attention-based framework for semantic image segmentation. Our approach is based on multi-class feature learning and learns the label pairs of the feature space given that each class is a vector of labels. We extend the supervised learning approach to perform segmentation by automatically learning the labels of labels, and then performing semantic segmentation as a step towards classification of labels in a shared feature space. Our approach improves both the classification and supervised learning performance of existing state-of-the-art semantic segmentation methods using only the label pairs. We demonstrate our approach for semantic segmentation and for image classification.
The goal of this paper is to propose a deep learning-based technique based on graph-structured graphical models for the optimization of optimal ranking. We propose to use Markov chain Monte Carlo (MCMC) to model the structure in graphs, and then a linear classifier based on the rank for each node, that achieves high ranking. We perform a regression analysis to evaluate the accuracy of our approach, which indicates that it converges significantly faster than the state-of-the-art method.
With increasing demands on neural network models, supervised object segmentation becomes increasingly important as well. In this work, we study the problem of automatically segmenting a dataset of object shapes in images and videos by means of supervised learning. We present a method that achieves very efficient segmentation of shapes, without sacrificing image quality or accuracy. For the sake of computational simplicity, we define a new set of supervised object segmentation criteria, and then develop an automatic segmentation algorithm for each shape.
Interpretable Machine Learning: A New Concept for Theory and Application to Derivative-Free MLPs
Deep End-to-End Neural Stacking
Multi-dimensional Recurrent Neural Networks for Music Genome Analysis
Recurrent Topic Models for Sequential Segmentation
A Convex Approach to Generalized Optimal RankingThe goal of this paper is to propose a deep learning-based technique based on graph-structured graphical models for the optimization of optimal ranking. We propose to use Markov chain Monte Carlo (MCMC) to model the structure in graphs, and then a linear classifier based on the rank for each node, that achieves high ranking. We perform a regression analysis to evaluate the accuracy of our approach, which indicates that it converges significantly faster than the state-of-the-art method.
With increasing demands on neural network models, supervised object segmentation becomes increasingly important as well. In this work, we study the problem of automatically segmenting a dataset of object shapes in images and videos by means of supervised learning. We present a method that achieves very efficient segmentation of shapes, without sacrificing image quality or accuracy. For the sake of computational simplicity, we define a new set of supervised object segmentation criteria, and then develop an automatic segmentation algorithm for each shape.