知の物理学研究センター / Institute for Physics of Intelligence (iπ)

【日時/Date】
2021年5月27日（木）10時30分～12時00分

【講演者/Speaker】
Nobuyuki YOSHIOKA/吉岡信行 （The University of Tokyo/東京大学）

【講演タイトル/Title】
“Encoding many-body quantum physics into neural networks”

【概要/Abstract】
The recent improvement of computational resources and optimization methods have enabled us to successfully apply neural networks to many machine learning tasks. It has been shown that neural networks are effective not only for classical data such as images and sounds, but also for representing the wavefunction of quantum many-body states [1].

In this talk, we introduce neural networks as variational wave functions for quantum many-body systems, and give an overview of their properties and applications.
The structure of the variational function, which does not rely on dimensionality, is complementary to the tensor networks used in large-scale calculations of quantum many-body systems. Computationally, neural networks can be optimized by the variational Monte Carlo method and achieve state-of-the-art accuracy, and analytically, it can represent many states (including topologically ordered states) exactly. We also discuss their application to non-equilibrium and finite-temperature properties [2, 3].

[1] G. Carleo and M. Troyer, Science 355, 602 (2017).
[2] N. Yoshioka and R. Hamazaki, Phys. Rev. B 99, 214306 (2019).
[3] Y. Nomura*, N. Yoshioka*, F. Nori, arXiv:2103.04791 (2021).

資料PDF/PDF

⇦ Top page