Learning User Preferences for Automated Question Answering – This thesis explores the use of word embeddings in machine learning to help identify the user’s emotional states (e.g. excitement or sadness) from the text of text. We demonstrate that this technique provides a powerful tool for identifying the emotional state that is associated with human emotional states in both text and visual data. Moreover, we argue that it leads to a significant gap between emotion-related content and the emotional state of a human being. We show how the use of emotion-related text can aid the identification of users’ emotional states in a variety of machine learning tasks such as sentiment analysis and emotion recognition. In particular, we illustrate how text-based emotion-related feature learning with the state-of-the-art neural network improves the robustness to human emotion detection and classification, and provides a new approach for generating emotions. We provide a comprehensive review of all previous work that has used emotion-related feature learning in emotion recognition.

The problem of learning a function from data is shown to be NP-hard to solve in the literature. While the answer to this problem has not yet been answered, the problem can be approached by exploiting an underlying information structure in the data. The data consist of data points sampled from several different distributions and the data is represented by a tree of points distributed over the distributions. The tree may contain nonlinear relationships as well as continuous ones. In this paper, we propose a supervised algorithm for learning the tree structure of data from the tree graphs. While this is a long time project, the proposal is still worth it as the proposed algorithm is significantly faster than the existing ones. The algorithm is designed to solve the sparsity problem under the existing supervised learning framework. We will show that the algorithm is competitive with a number of supervised learning approaches on a simulated and real world data set.

Artificial neural networks for predicting winter weather patterns on maps of Europe

TBD: Typed Models

Learning User Preferences for Automated Question Answering

Improving the Robustness of Deep Neural Networks by Exploiting Connectionist Sampling

On the Convergence of Sparsity Regularization for the Prediction of Gene Expression VariantsThe problem of learning a function from data is shown to be NP-hard to solve in the literature. While the answer to this problem has not yet been answered, the problem can be approached by exploiting an underlying information structure in the data. The data consist of data points sampled from several different distributions and the data is represented by a tree of points distributed over the distributions. The tree may contain nonlinear relationships as well as continuous ones. In this paper, we propose a supervised algorithm for learning the tree structure of data from the tree graphs. While this is a long time project, the proposal is still worth it as the proposed algorithm is significantly faster than the existing ones. The algorithm is designed to solve the sparsity problem under the existing supervised learning framework. We will show that the algorithm is competitive with a number of supervised learning approaches on a simulated and real world data set.