Seminario 20/25: Dolores Romero Morales (Copenhagen Business School, Denmark): Explaining Machine Learning Outcomes by Means of Mathematical Optimization

Información

There is a growing literature on enhancing the interpretability of Machine Learning methods involved in Data Driven Decision Making. Interpretability is desirable for non-experts; it is required by regulators for models aiding, for instance, credit scoring; and since 2018 the European Union has extended this requirement by imposing the so-calledright-to-explanation in algorithmic decision making. Mathematical Optimization has shown a crucial role when striking a balance between interpretability and accuracy, having LASSO as one of the main exponents. In this presentation, we will go through very recent examples of enhancing the interpretability of Supervised as well as Unsupervised Learning methods with the help of Mixed Integer NonLinear Optimization.