AI Paper Spotlights - Curve your Enthusiasm: Making “interpretable ML” more interpretable
In this talk, our ML Researcher, Alma Lindborg, presents the paper “Curve Your Enthusiasm: Concurvity Regularization in Differentiable Generalized Additive Models”. This work focuses on Generalized Additive Models (GAMs), a class of machine learning models that is particularly favored where a high degree of accountability is required for model decisions, such as in hiring processes or the medical domain. Although it is easy to visualize how each feature (e.g., age, blood test values) contributes to a prediction (e.g., illness) in these models, we show that these interpretations can be misleading due to concurvity, i.e., mutual dependencies between the features. We show that by regularizing concavity, we can constrain the model to find a more consistent relationship between features and prediction, thus creating a more robust interpretation.