Ukrainian Catholic University (UCU) launched, in collaboration with the Center for Responsible AI at NYU, a long-awaited series of meet-ups on the topic of Responsible AI.

The goal was to invite individuals who contribute to the advancement of responsible AI development and its practical applications. The first meeting was led by Denys Herasymuk, one of the first research fellows in the #RAIforUkraine program. He shared his personal experiences along with the findings from his research, which he conducted in collaboration with his mentors.

Responsible Approaches to Missing Value Imputation and Model Selection in Machine Learning

Abstract: The talk covers my experience in the “RAI for Ukraine” program and highlights our research on responsible AI, focusing on advancements aimed at ensuring the stability and fairness of machine learning models. I will begin by sharing key insights and accomplishments from my time in the program, followed by a brief introduction to model stability and uncertainty quantification. Next, I will highlight the challenges of incorporating these dimensions of model performance into the model selection process. Building on this foundation, I will introduce our comprehensive software library for model auditing and responsible model selection, called Virny, along with an interactive tool called VirnyView. Finally, I will present our Shades-of-Null benchmark for responsible missing value imputation. This large-scale empirical study, involving 23,940 experimental pipelines, provides a comprehensive and rigorous evaluation of various families of imputation methods on a wide range of evaluation metrics, in plausible and socially meaningful missingness scenarios.

The talk is based on two papers:

A big thank you to Tetiana Zakharchenko for co-hosting and organizing the event!

The next meet-up is set for November 28 at 6:00 PM in Kyiv, and it will feature Mariana Romanyshyn, Area Tech Lead at Grammarly. Stay tuned for further details!