Mitigating Social Challenges among Vulnerable Communities with Machine Learning
Open Access
- Author:
- Tabar, Maryam
- Graduate Program:
- Informatics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 16, 2023
- Committee Members:
- Jeffrey Bardzell, Program Head/Chair
Dongwon Lee, Co-Chair & Dissertation Advisor
Amulya Yadav, Co-Chair & Dissertation Advisor
Carleen Maitland, Major Field Member
C Lee Giles, Major Field Member
Rayid Ghani, Special Member
S. Shyam Sundar, Outside Unit & Field Member - Keywords:
- Artificial Intelligence/Data Science for Social Good
Machine Learning - Abstract:
- There are various environmental and social challenges that disproportionately affect vulnerable communities in society. Extensive research has been conducted in various fields, such as agricultural sciences and social sciences, to understand some of those challenges and design intervention/prevention programs. However, effective/efficient implementation of mitigation plans is usually highly challenging in the field. Inspired by recent advances in Machine Learning (ML), this dissertation mainly focuses on the adaptation of ML-based techniques in certain real-world domains under various challenges to help address several social problems in a more effective/efficient manner. In fact, it focuses on two real-world domains, AI for Agriculture and AI for Social Welfare of Housing-Insecure Low-Income Americans, and addresses some challenges by proposing solutions tailored to the characteristics of the motivating problem domain. For example, to address the challenge of a lack of ground-truth labels, it proposes a label generation approach that translates the findings of social science research to high-quality labels to facilitate training ML models. Additionally, it proposes a loss function to improve the learning of neural networks when only coarse-grained ground-truth labels are available. In conclusion, this dissertation aims to adapt ML algorithms in specific real-world domains with particular challenges and characteristics.