Multitask Learning with Learned Semantic-Aware Hierarchical Representations – (JNU 2017) The use of semantic knowledge for intelligent systems is an emerging field in computer-human-computer interaction. While it has recently received increasing attention, it is still largely unexplored. In this paper, we propose a new method to solve it by solving an end-to-end system learning problem under natural language models. Specifically, we first develop a stochastic nonnegative matrix factorization framework to handle the semantic-aware-memory problem. To this end, we first propose an adaptive learning algorithm to solve the semantic-aware-memory problem, which is then augmented with a learning matrix factorizer. Finally, we propose a nonnegative matrix factorization algorithm for solving the semantic-aware-memory problem that allows for the efficient use of the semantic-aware-memory model. Our algorithms are particularly applicable for solving the semantic-aware-learning-problem and have been compared to state-of-the-art learning algorithms on two benchmark datasets.
The concept of non-monotonic decision-making is a crucial property that distinguishes different domains of decision-making and provides rich explanations for the phenomena. In this paper we first present a model of this property, showing how it relates to a probabilistic approach to decision-making. The model was designed with such a perspective, that we can compare one domain to the other and to the different types of decision-making possible. Then we develop a framework for an interactive approach to non-monotonic decision-making. We present a new methodology for the construction of models which can be used to learn the relationships between different domains. In particular, we present a first model which can automatically learn the relationship between domain distributions. The proposed approach is the first to develop a probabilistic approach to non-monotonic decision-making in a complex decision-making environment. We demonstrate the method on a real world dataset and demonstrate that it performs well over a classification rule that could have been used to categorize the data.
Augmenting Web Page Visibility Dataset with Disparate Linguistic Attention
Multitask Learning with Learned Semantic-Aware Hierarchical Representations
Visual Tracking via Superpositional Matching
Towards Understanding and Explaining the Decision Making in Complex DomainsThe concept of non-monotonic decision-making is a crucial property that distinguishes different domains of decision-making and provides rich explanations for the phenomena. In this paper we first present a model of this property, showing how it relates to a probabilistic approach to decision-making. The model was designed with such a perspective, that we can compare one domain to the other and to the different types of decision-making possible. Then we develop a framework for an interactive approach to non-monotonic decision-making. We present a new methodology for the construction of models which can be used to learn the relationships between different domains. In particular, we present a first model which can automatically learn the relationship between domain distributions. The proposed approach is the first to develop a probabilistic approach to non-monotonic decision-making in a complex decision-making environment. We demonstrate the method on a real world dataset and demonstrate that it performs well over a classification rule that could have been used to categorize the data.