Generating Semantic Representations using Greedy Methods – This paper presents an end-to-end approach for the semantic parsing of human language using machine learning based tools. The proposed approach is composed of two steps: (i) it learns semantic relationships by learning the representations learned by the human experts and (ii) it provides a natural way for semantic parsers to interact with the human experts. The proposed end-to-end approach is evaluated and compared with a previous state-of-the-art baseline based on Semantic Pattern Recognition (SPR) tasks. The results show that our approach outperforms the other baseline models compared to human experts without sacrificing performance by a large degree.

In this paper, we present LBP, a new framework for real-time multi-label classification, in which a real-time model is trained by a supervised machine learning based feed-forward Neural Network with a mixture of Convolutional Neural Network (CNN), which learns a mixed bag of labels to classify multiple labels and labels to classify multiple label samples. We study the importance of a training set for LBP. In our study, we present a novel training network architecture to directly train a multi-label classifier. We present two general-purpose features that help the new approach: the CNN model in terms of the feature space to be trained, and each network in terms of its specific task, which are learned through learning a joint model from all the labels to a single, globally distributed label. Based on these features, LBP can learn and classify multiple labels. Experiments on both synthetic and real data sets confirm the effectiveness of LBP for both training and learning tasks.

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

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

Generating Semantic Representations using Greedy Methods

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

Multi-label Multi-Labelled Learning for High-Dimensional Data: A Meta-StudyIn this paper, we present LBP, a new framework for real-time multi-label classification, in which a real-time model is trained by a supervised machine learning based feed-forward Neural Network with a mixture of Convolutional Neural Network (CNN), which learns a mixed bag of labels to classify multiple labels and labels to classify multiple label samples. We study the importance of a training set for LBP. In our study, we present a novel training network architecture to directly train a multi-label classifier. We present two general-purpose features that help the new approach: the CNN model in terms of the feature space to be trained, and each network in terms of its specific task, which are learned through learning a joint model from all the labels to a single, globally distributed label. Based on these features, LBP can learn and classify multiple labels. Experiments on both synthetic and real data sets confirm the effectiveness of LBP for both training and learning tasks.