Summary
2023
Session Number:A3L-4
Session:
Number:A3L-42
New Learning Algorithm of Gaussian–Bernoulli Restricted Boltzmann Machine and its Application in Feature Extraction
Yasuda Muneki, Xiong Zhongren,
pp.134-137
Publication Date:2023-09-21
Online ISSN:2188-5079
Summary:
Feature extraction is essential in various data analyses applications. Therefore, the development of an universal feature extractor is critical. A restricted Boltzmann machine (RBM) is a powerful candidate for such an universal feature extractor. In this study, we focus on a Gaussian--Bernoulli RBM (GBRBM) and canonicalize it through re-parameterization. An effective learning algorithm for the canonicalized GBRBM is proposed based on spatial Monte Carlo integration. Using numerical experiments, we demonstrate that the GBRBM outperforms the standard denoising autoencoder as the feature extractor in strong noisy environments.