From Finance to Space
A recent publication by Emanuele Borgonovo and Giuseppe Savaré (both of Bocconi University’s Department of Decision Sciences) with Alessio Figalli (ETH Zürich, and the 2018 Fields Medal) and Elmar Plischke (Clausthal University of Technology), published in Management Science, introduces an innovative approach for finding the important variables in complex machine learning and scientific models.
The approach is based on the theory of optimal transport, a mathematical theory originated by Gaspard Monge in 1781, that provides a powerful framework to define the new importance measures. Thanks to the solid structure, the sensitivity measures possess relevant properties such as zero-independence, max-functionality and monotonicity, which are regarded as crucial in the recent literature on explainable Artificial Intelligence and Data Science.
To ensure their method is practical, the authors apply it in a real-world supply chain scenario, analyzing an assemble-to-order system, a crucial process in industries like manufacturing and retail. This new method, however, is not just for logistics. It helps decision-makers across various industries where complex models are deployed, for instance:
- In space mission planning, in a work in cooperation with NASA for the MARS sample return program;
- In climate modeling, where it helps policy makers by pinpointing the most critical factors affecting long-term predictions;
- In healthcare, it can enhance understanding of how different treatment variables impact patient outcomes;
- In finance, it can help investors identify the most influential risk factors in portfolio performance.
