Speakers: Azim Ahmadzadeh, PhD

Azim Ahmadzadeh, PhD

Assistant Professor
Department of Computer Science
University of Missouri-St. Louis

 

Dr. Ahmadzadeh is a computer scientist, a data scientist, and an interdisciplinary researcher. His research primarily focuses on the effective use of Artificial Intelligence (Machine Learning and Deep Neural Networks) to provide physicists with a better understanding of our closest star, the Sun, and the physical processes driving its activities.

 

 

Unifying Framework for Comparing “Apples to Oranges”

Evaluating machine learning (ML) models fairly is often challenging due to differences in training and validation settings. In this talk, I will introduce Contingency Space, a novel evaluation framework that bridges this gap by providing a more comprehensive comparison of models, without complicating the task. Contingency Space contextualizes any preferred evaluation measure (e.g., precision, recall, F-score, etc.) by representing a model’s performance within the joint distribution of true-positive and true-negative rates. This approach turns a measure into a surface and uncovers performance insights that are often obscured by metric-specific biases. Furthermore, leveraging the geometrical properties in this space allows for a critical assessment of each metric’s effectiveness in relation to specific objectives, as well as an evaluation of their strengths and weaknesses. Finally, I will introduce a Python package that implements this framework, enabling researchers to integrate it seamlessly into their work, regardless of their preferred evaluation metrics.