Learning Fairness and Graph Deep Generation in Dynamic Environments
Loading...
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
2020-01-20
Type of Work
Department
Information Systems
Program
Information Systems
Citation of Original Publication
Rights
Distribution Rights granted to UMBC by the author.
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Subjects
Abstract
Machine learning systems are employed nowadays in an ever growing number of applications and have become a necessity in all aspects of our lives. Compared with machine learning approaches designed to work in static environments, dynamic models are also of great importance with wide applicability in a plethora of real-world applications, ranging from IoT analytics, screening applicants to simulation studies. In the meanwhile, the dynamic property of environments also brings unique and complicated challenges such as data is continually entering the system and the information source might also change over time. In this dissertations, we have broadly explored fairness-aware learning and deep graph generation in such challenging dynamic environments, and make several advances for supporting the practical development and deployment of applications that are dynamic in nature. The first contribution is to develop two online decision trees with fairness to support social fairness. Specially, we propose the new fair splitting criteria consider both data encoding and discrimination elimination, and additionally take evolving data distribution into consideration along with an effectively and robust trade-off between prediction accuracy and fairness performance. Experimental results show that the proposed algorithms are able to deal with discrimination in massive and non-stationary streaming environments, with a better trade-off between fairness and predictive performance. Continuing the theme of learning online with fairness, we next propose and illustrate a flexible ensemble model for fair online decision-making. The conducted experiments show that the proposed model improves fairness dramatically while maintaining a fairly comparable predictive performance. This study also theoretically analyzes the inadequacy of current sampling approaches in fairness literatures and introduces a new effective sampling direction with experimental verification. Finally, deep generative models for graphs have exhibited promising performance in ever increasing domains such as design of molecules (i.e, graph of atoms) and structure prediction of proteins (i.e., graph of amino acids). Existing work is on static rather than dynamic graphs and their application to protein folding, molecule reactions, and human mobility. We propose a generic framework of deep generative models for interpretable dynamic graph generation, providing a necessary complement to this important yet challenging task. The experimental evaluation results show the flexibility and versatility of the proposed models in generating dynamic graphs.