Café com Física

Machine Learning (ML) in Physics

by Tuhin Malik (CFisUC)

AD.1 (Departamento de Física FCTUC)


Departamento de Física FCTUC

Universidade de Coimbra

Machine Learning (ML) is a field with great success for many practical applications and active research topics. In recent years, ML algorithms are being widely used in fundamental research, and physics is no exception for converting information into knowledge. The ML provides a powerful tool to classify and predict patterns, even in complex data sets.
The ML could be a powerful tool to extract information on missing theory in physics.

The main drawback of ML is overfitting and the overfitted model may predict something very crazy. However, there is some standard cross-check for that. The talk will start from the very basics of ML. I will also discuss how to implement it in a Physics problem with great care.  The randomization of testing will be discussed, to further reduce the risk of accidentally producing a good result.

To have a theoretical understanding of any system, a physics-based model is necessary. ML algorithms cannot replace physics modeling in that respect. However, an interesting area of future work might be in combining the theoretical model and the machine-learning methods to arrive at a better physical model.

ID da reunião: 81310717360
Password: 055056


Organized by

Filipe Veloso e Pedro Costa