Skip to main content

Mitigating Bias with Data Science, Machine Learning, and AI

IMPACT intern Melanie Sharif combined philosophic, probabilistic, psychological, and computer science approaches to research mitigating bias in machine learning and artificial intelligence.

Every year, IMPACT welcomes interns who provide skills and assistance to project teams while gaining work experience. This summer, under the guidance of Dr. Steve Crawford, Dr. Manil Maskey, and Iksha Gurung, intern Melanie Sharif researched mitigating bias in machine learning and artificial intelligence. Melanie is a graduate student at the University of Colorado in Boulder studying creative technology and design with a focus on human-computer interaction.

Coming from a background in psychology, Melanie says mitigating bias is an interdisciplinary effort, combining philosophic, probabilistic, psychological, and computer science approaches. While it has a primarily psychological undertone, it is also of particular interest for computer scientists and policy makers. She notes:

"Bias has broad impacts across all scientific fields and disciplines, such as in AI ethics for science, bias-informed results, smarter pipelines, and the promotion of science-informed decision making on a policy level."

Four orange pie charts demonstrate imbalance bias
Image Caption

An example of imbalance bias visualization for an ML model

Melanie found that researchers can get insights into bias through data visualization, a tool in a ML practitioner’s “toolbelt” that they can use to understand their data and their models. She also studied more direct methods such as rebalancing algorithms, using premade algorithms with various methods or approaches and results that practitioners can apply to their work to make their results more fair. An example of this is SMOTE (Fair-Synthetic Minority Oversampling Technique) which rebalances positive and negative class designations and removes biased labels in a given training data set. Other tools she used include Google Facets, IBM AI Fairness 360, and Fairkit-learn. She suggests future research could focus on developing new quantification methods, exploring other ways to use visualization, and assessing ongoing policy development.

This summer’s research aligns well with Melanie’s previous experience and future goals. Before starting her M.S. program, she was an AmeriCorps member and a Math Fellow at Denver Public Schools, working with Title IV schools in the Denver area to improve educational outcomes. In the future, she endeavors to do ML engineering work at a firm that aligns with her values, and/or pursue a PhD in an AI-related field. Following that, she hopes to return to the agency to expand on her work this past summer and potentially work on space exploration projects.

Check out Melanie's LinkedIn profile.

More information about IMPACT can be found at NASA Earthdata and the IMPACT project website.

Details

Last Updated

Published