Data: 11 de dezembro de 2023
Horário: 14 horas
Local: Sala 2077 do prédio do ICEx
Title: Interpretable Machine Learning for Health Informatics
Abstract: In model-based clinical decision support, clinicians must be able independently verify the recommendations given by models to responsibly care for patients. To this end, the field of explainable AI has invested considerable attention to explaining black-box models with post-hoc explanations such as Shapley values, with mixed results. In this talk I will discuss an alternative approach – using machine learning to discover interpretable models explicitly – and discuss the pros and cons of this approach. I will present a case study of using white-box machine learning, a variant of symbolic regression, to generate interpretable models for risk of secondary hypertension.
