2019Journal of Advanced Computational Intelligence and Intelligent InformaticsOpen access

Crystalizing Effect of Simulated Annealing on Boltzmann Machine

Hiroki Shibata, Hiroshi Ishikawa, Yasufumi Takama

Open full text 2 citations

Abstract

This paper proposes a method to estimate the posterior distribution of a Boltzmann machine. Due to high feature extraction ability, a Boltzmann machine is often used for both of supervised and unsupervised learning. It is expected to be suitable for multimodal data because of its bi-directional connection property. However, it needs a sampling method to estimate the posterior distribution, which becomes a problem during an inference period because of the computation time and instability. Therefore, it is usually converted to feedforward neural networks, which means to lose its bi-directional property. To deal with these problems, this paper proposes a method to estimate the posterior distribution of a Boltzmann machine fast and stably without converting it to feedforward neural networks. The key idea of the proposed method is to estimate the posterior distribution using a simulated annealing on non-uniform temperature distribution. The advantage of the proposed method against Gibbs sampling and conventional simulated annealing is shown through experiments with artificial dataset and MNIST. Furthermore, this paper also gives the mathematical analysis of Boltzmann machine’s behaviour with regard to temperature distribution.

About this research paper

What this paper is about

This paper proposes a method to estimate the posterior distribution of a Boltzmann machine. Due to high feature extraction ability, a Boltzmann machine is often used for both of supervised and unsupervised learning. It is expected to be suitable for multimodal data because of its bi-directional connection property. However, it needs a sampling method to estimate the posterior distribution, which becomes a problem during an inference period because of the computation time and instability. Therefore, it is usually converted to feedforward neural networks, which means to lose its bi-directional property. To deal with these problems, this paper proposes a method to estimate the posterior distribution of a Boltzmann machine fast and stably without converting it to feedforward neural networks. The key idea of the proposed method is to estimate the posterior distribution using a simulated annealing on non-uniform temperature distribution. The advantage of the proposed method against Gibbs sampling and conventional simulated annealing is shown through experiments with artificial dataset and MNIST. Furthermore, this paper also gives the mathematical analysis of Boltzmann machine’s behaviour with regard to temperature distribution.

Why it matters

OpenAlex reports 2 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

This paper proposes a method to estimate the posterior distribution of a Boltzmann machine. Due to high feature extraction ability, a Boltzmann machine is often used for both of supervised and unsupervised learning. It is expected to be suitable for multimodal data because of its bi-directional connection property. However, it needs a sampling method to estimate the posterior distribution, which becomes a problem during an inference period because of the computation time and instability. Therefore, it is usually converted to feedforward neural networks, which means to lose its bi-directional property. To deal with these problems, this paper proposes a method to estimate the posterior distribution of a Boltzmann machine fast and stably without converting it to feedforward neural networks. The key idea of the proposed method is to estimate the posterior distribution using a simulated annealing on non-uniform temperature distribution. The advantage of the proposed method against Gibbs sampling and conventional simulated annealing is shown through experiments with artificial dataset and MNIST. Furthermore, this paper also gives the mathematical analysis of Boltzmann machine’s behaviour with regard to temperature distribution.

Key concepts: Boltzmann machine, Computer science, Boltzmann distribution, Restricted Boltzmann machine, Posterior probability, MNIST database, Simulated annealing, Computation

Related papers

Back to paper searchBrowse research topicsOriginal source
Crystalizing Effect of Simulated Annealing on Boltzmann Machine — Research Paper | ScholarLens