Yuval Moskovitch

University of Michigan

Position: Postdoctoral Researcher
Rising Stars year of participation: 2021
Bio

Yuval is a postdoctoral researcher in the Database Research Group at the University of Michigan hosted by Prof. H. V. Jagadish. Her research is centered around data management advanced database applications provenance and process analysis. In her current research she focuses on data management for fairness and responsible data science. Yuval obtained her Ph.D. in Computer Science from Tel Aviv University under the supervision of Prof. Daniel Deutch. She completed a BSc in Software Engineering and MSc in Computer Science at Ben Gurion University. Yuval is the recipient of several awards including the Shulamit Aloni Scholarship for Advancing Women in Science by the Israeli Ministry of Science and Technology and the Data Science Fellowship for outstanding postdocs of the planning and budgeting committee of the council for higher education.

Towards Reliable Data-Driven Decision Tools

Towards Reliable Data-Driven Decision Tools
The ubiquity of data in recent years has led to wide use of automated data-driven decision-making tools. These tools are gradually supplanting humans in a vast range of application domains from deciding who should get a loan to automated hiring student grading and even in assessing the risk of paroling convicted criminals. The growing dependence we developed on these tools in particular in domains where data-driven algorithmic decision-making may affect human life raises concerns regarding their reliability. Indeed with the increasing use of data-driven tools we also witness a large number of cases where these tools are biased or discriminate unfairly. In light of the reported cases users have reason to worry about the results produced by these tools. The research question then is how to make these tools more reliable. The need to address such concerns has been felt by many. There is a significant body of literature on fairness in AI on estimating the reliability of statements representing a document or dataset and so on. However of necessity most such works have addressed a particular narrow concern. The challenge for a data science pipeline is that it has to establish end-to-end reliability. In my research I develop methods that intervene in the development pipeline of data-driven systems at different points from the data collection phase through the data analysis and the assessment of the Machine Learning model. The proposed methods allow users to tackle various sources of bias by determining fitness-for-use of data before the model training analyzing the performance of the model to detect groups that are treated unfairly and assessing the quality of the results of aggregation queries which are commonly used by analysts throughout the development of data-driven decision tools.