Speaker

Prof. Akio Tomiya
International Professional University of Technology in Osaka

Title

Equivariant Transformer is All You Need

Abstract

The simulation of quantum systems has been transformed by machine learning, with equivariance playing a crucial role to achieve good efficiency. However, previous methods have faced challenges, particularly in self-learning Monte Carlo (SLMC) simulations, leading to poor acceptance rates.In this seminar, we introduce a symmetry equivariant attention mechanism for SLMC, designed to overcome these limitations. Our approach exhibits a scaling law similar to large language models applied to a spin-fermion model on a two-dimensional lattice. It improves model quality with the number of layers.I assume that for a physicist audience without machine learning expertise, the seminar will provide an accessible introduction to machine learning and equivariance. We will then explain in detail our new method, emphasizing its potential to develop more accurate and efficient Monte Carlo algorithms for simulating physical systems.

Join Zoom Meeting

Meeting ID: 921 8452 8527
Passcode: 206283