Metrics
390 File views/ downloads
111 Record Views
Dissertation
Machine Learning Guided by Linguistic and Behavioral Knowledge
University of West Florida Libraries
Doctor of Philosophy (PHD), University of West Florida
2021
Abstract
The recent success of AI has been primarily driven by the extraordinary progress in the field of machine learning. The ultimate goal of machine learning is to develop algorithms capable of making accurate predictions in an explainable way by learning efficiently from a small amount of training data. Despite an exceptionally fast-paced growth, machine learning has been exceedingly successful in achieving accurate predictions, at the cost of sacrificing most of, if not all, explainability and by relying on huge amount of training data. Recent work has, on the other hand, shown that domain knowledge, when properly incorporated in learning algorithms, can facilitate learning from small data sets and provide various forms of explainability. In this dissertation, I propose novel ways of incorporating linguistic and behavioral knowledge into machine learning models for achieving different goals such as improving prediction accuracy, using less data, increase explainability, and evaluating cognitive biases. We exemplify our novel approaches on some challenging tasks that require special treatment either due to lack of data and/or need for explainable predictions. We first consider extracting spatial relations from language, which is a complex task due to the ambiguity of spatial relations and scarcity of available training data. To this end, we use linguistic knowledge to define various constraints imposed on classifiers to infer the correct classifications holistically. Human choice prediction is the other domain that we consider because of the fundamental role it plays in the understanding of human behavior and in the design of intelligent systems. We propose novel methods to leverage procedural knowledge, in the form of psychological models of decision making, in combination with machine learning, to achieve better predictions, understand the underlying deliberation processes, and elicit user preferences. Finally, we extend our work to the domain of sequential decision making by designing agents that learn constraints from demonstrations and then use cognitive models as orchestrators to exploit these learned constraints for making choices between conflicting goals. We use various real world and synthetic data to evaluate our proposed methods throughout this dissertation. Our experimental results show the efficacy of our methods which significantly improves upon the state-of-the-art in all of the considered tasks.
Details
- Title
- Machine Learning Guided by Linguistic and Behavioral Knowledge
- Resource Type
- Dissertation
- Publisher
- University of West Florida Libraries; [Pensacola, Florida] :
- Format
- pdf
- Number of pages
- 154
- Copyright
- 9.94E+16
- Identifiers
- 99380090853706600
- Academic Unit
- Intelligent Systems and Robotics; School of Education
- Language
- English
- Awarding Institution
- University of West Florida; Doctor of Philosophy (PHD)
- Theses and Dissertations
- Doctor of Philosophy (PHD), University of West Florida