A Model of Amoeba-Based Neurocomputer

Aono, Masashi; Hirata, Yoshito; Hara, Masahiko; Aihara, Kazuyuki

doi:10.2477/jccj.h2119

Cited by 4 publications

(2 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By extracting the essential dynamics of the multilevel optimization processes [14,15], the computing scheme may be implemented by other faster materials capable of multilevel self-organization. Placing all the unique features together, the proposed amoeba-based system may open up completely new methods of knowledge discovery and data mining.…”

Section: Knowledge Discovery By Means Of Autonomous Meta-problem Smentioning

confidence: 99%

Amoeba-Based Knowledge Discovery System

Munakata

Aono

Hara

2010

2010 Third International Joint Conference on Computational Science and Optimization

Self Cite

View full text Add to dashboard Cite

We propose an amoeba-based knowledge discovery or data mining system, that is implemented using an amoeboid organism and an associated control system. The amoeba system can be considered as one of the new non-traditional computing paradigms, and it can perform intriguing, massively parallel computing that utilizes the chaotic behavior of the amoeba. Our system is a hybrid of a traditional knowledge-based unit implemented on an ordinary computer and an amoeba-based search unit, with an interface of an optical control unit. The solutions in our system can have one-to-one mapping to solutions of other well known areas such as neural networks and genetic algorithms. This mapping feature allows the amoeba to use and apply techniques developed in other areas. Various forms of knowledge discovery processes are introduced. Also, a new type of knowledge discovery technique, called "autonomous meta-problem solving," is discussed.

show abstract

Section: Knowledge Discovery By Means Of Autonomous Meta-problem Smentioning

confidence: 99%

Amoeba-Based Knowledge Discovery System

Munakata

Aono

Hara

2010

2010 Third International Joint Conference on Computational Science and Optimization

Self Cite

View full text Add to dashboard Cite

show abstract

“…More interesting and important processes in chemistry, physics and biology take place on time scales much longer than what is accessible in the molecular dynamics (MD) and/or Monte Carlo (MC) simulations. For example, in studying various natural events as proteins folding, [1,2] chemical reactions, [3] and polymorphic transformation in solids, [4] high free energy barriers must be surmounted, which is unlikely to occur within the time scales available in brute force molecular simulations. Over the past decades, many successful techniques have been developed to sample the full configuration space within the simulation times accessible to present-day molecular simulation methods.…”

Section: Introductionmentioning

confidence: 99%

Adaptive‐numerical‐bias metadynamics

2017

View full text Add to dashboard Cite

A metadynamics scheme is presented in which the free energy surface is filled with progressively adding adaptive biasing potentials, obtained from the accumulated probability distribution of the collective variables. Instead of adding Gaussians with assigned height and width in conventional metadynamics method, here we add a more realistic adaptive biasing potential to the Hamiltonian of the system. The shape of the adaptive biasing potential is adjusted on the fly by sampling over the visited states. As the top of the barrier is approached, the biasing potentials become wider. This decreases the problem of trapping the system in the niches, introduced by the addition of Gaussians of fixed height in metadynamics. Our results for the free energy profiles of three test systems show that this method is more accurate and converges more quickly than the conventional metadynamics, and is quite comparable (in accuracy and convergence rate) with the well-tempered metadynamics method. © 2017 Wiley Periodicals, Inc.

show abstract