Veronica Guidetti (Unimore)

Can symbolic machine learning catalyze science?

In recent years, machine learning has experienced a surge in capacity due to increasingly complex algorithms, particularly in deep learning. While deep learning excels in handling complex data, it sacrifices model interpretability. This talk aims to present and promote symbolic machine learning as an alternative, offering understandable models for high-risk and scientific domains. Symbolic approaches are crucial in fields such as applied mathematics and physics to combine new data-driven knowledge with existing theories. We will discuss the building blocks of symbolic regression methods, their evolution over time, and their achievements. Additionally, we will present recent applications in theoretical physics, focusing particularly on cosmological inflation models, and discuss future lines of research.