14–15 Dec 2023
Centre for Particle Theory
Europe/London timezone

Towards interpretable machine learning with Restricted Boltzmann Machine

14 Dec 2023, 17:50
20m
Ph30 (Centre for Particle Theory)

Ph30

Centre for Particle Theory

Long Talk (20 mins) Lattice Physics

Speaker

Chanju Park (Swansea University)

Description

Restricted Boltzmann Machines (RBMs) are energy-based machine learning models inspired by the spin-glass system. Using its energy function, we can analyse its training dynamics in terms of statistical mechanics and map its hyperparameters to thermodynamic quantities. This interpretability unravels the black-box nature of the machine-learning process and allows us to understand the underlying mechanisms.

Are you happy for your talk to be recorded? Yes
Please select the most relevant category Machine Learning
Would you be interested in receiving feedback on your presentation? Yes

Primary author

Chanju Park (Swansea University)

Co-authors

Biagio Lucini (Swansea University) Gert Aarts (Swansea University)

Presentation materials