Innovative machine learning algorithm driven by violation of detailed balance condition

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K13849
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Mathematical physics/Fundamental condensed matter physics
• Research Institution: Tohoku University
• Principal Investigator: Ohzeki Masayuki 東北大学, 情報科学研究科, 准教授 (80447549)
• Co-Investigator(Kenkyū-buntansha): 一木 輝久 名古屋大学, 未来社会創造機構, 特任准教授 (40711156)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 詳細釣り合い / 古典確率 / 量子ゆらぎ / 機械学習 / 汎化性能

Outline of Final Research Achievements

This study is to develop an innovative algorithm in the field of machine learning that uses classical stochastic processes, based on the fact that the convergence to steady state accelerates when the detailed balance is broken.
Starting with Physical Review E 93 (2016) 012129 in 2016, we understand the role played by detailed balance as a physical process, go beyond classical stochastic processes, and step into quantum stochastic processes to create diverse algorithms I aimed at. The basic theory for exploiting quantum systems that are difficult to handle due to the presence of the negative sign problem was developed in Scientific Reports, (2017) 41186, and in scientific reports, 8 (2018) 9950, machine learning utilizing quantum fluctuation A demonstration experiment of the algorithm was conducted.

Free Research Field

機械学習、量子力学、統計力学

Academic Significance and Societal Importance of the Research Achievements

機械学習の発展は現代科学の礎を築く上で最重要課題であり、その物理学的視座に基づく新規アルゴリズム創出は、決して発見論的ではなく、検証可能であり確固たる理論体系の元に築き上げられる。場当たり的な手法ではない普遍的な手法となるため、その構造の理解と手法の水平展開の容易さから、今後10年に渡る研究の展開の起点となることが期待される。