TY - JOUR

T1 - Decision maker based on atomic switches

AU - Kim, Song Ju

AU - Tsuruoka, Tohru

AU - Hasegawa, Tsuyoshi

AU - Aono, Masashi

AU - Terabe, Kazuya

AU - Aono, Masakazu

N1 - Publisher Copyright:
© 2016, Song-Ju Kim, et al.

PY - 2016

Y1 - 2016

N2 - We propose a simple model for an atomic switch-based decision maker (ASDM), and show that, as long as its total number of metal atoms is conserved when coupled with suitable operations, an atomic switch system provides a sophisticated "decision-making" capability that is known to be one of the most important intellectual abilities in human beings. We considered a popular decisionmaking problem studied in the context of reinforcement learning, the multi-armed bandit problem (MAB); the problem of finding, as accurately and quickly as possible, the most profitable option from a set of options that gives stochastic rewards. These decisions are made as dictated by each volume of precipitated metal atoms, which is moved in a manner similar to the fluctuations of a rigid body in a tug-of-war game. The "tug-of-war (TOW) dynamics" of the ASDM exhibits higher efficiency than conventional reinforcement-learning algorithms. We show analytical calculations that validate the statistical reasons for the ASDM to produce such high performance, despite its simplicity. Efficient MAB solvers are useful for many practical applications, because MAB abstracts a variety of decisionmaking problems in real-world situations where an efficient trial-and-error is required. The proposed scheme will open up a new direction in physics-based analog-computing paradigms, which will include such things as "intelligent nanodevices" based on self-judgment.

AB - We propose a simple model for an atomic switch-based decision maker (ASDM), and show that, as long as its total number of metal atoms is conserved when coupled with suitable operations, an atomic switch system provides a sophisticated "decision-making" capability that is known to be one of the most important intellectual abilities in human beings. We considered a popular decisionmaking problem studied in the context of reinforcement learning, the multi-armed bandit problem (MAB); the problem of finding, as accurately and quickly as possible, the most profitable option from a set of options that gives stochastic rewards. These decisions are made as dictated by each volume of precipitated metal atoms, which is moved in a manner similar to the fluctuations of a rigid body in a tug-of-war game. The "tug-of-war (TOW) dynamics" of the ASDM exhibits higher efficiency than conventional reinforcement-learning algorithms. We show analytical calculations that validate the statistical reasons for the ASDM to produce such high performance, despite its simplicity. Efficient MAB solvers are useful for many practical applications, because MAB abstracts a variety of decisionmaking problems in real-world situations where an efficient trial-and-error is required. The proposed scheme will open up a new direction in physics-based analog-computing paradigms, which will include such things as "intelligent nanodevices" based on self-judgment.

KW - Amoeba-inspired computing

KW - Atomic switch

KW - Multi-armed bandit problem

KW - Natural computing

KW - Reinforcement learning

KW - Tug-of-war dynamics

UR - http://www.scopus.com/inward/record.url?scp=84979584926&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84979584926&partnerID=8YFLogxK

U2 - 10.3934/matersci.2016.1.245

DO - 10.3934/matersci.2016.1.245

M3 - Article

AN - SCOPUS:84979584926

SN - 2372-0484

VL - 3

SP - 245

EP - 259

JO - AIMS Materials Science

JF - AIMS Materials Science

IS - 1

ER -