A review on generative Boltzmann networks applied to dynamic systems. (15th January 2021)

• Record Type: Journal Article
• Authors: Souriau, Rémi; Lerbet, Jean; Chen, Hsin; Vigneron, Vincent
• Is Part Of: Mechanical systems and signal processing. Volume 147(2021)
• Publication Date: 2021-01-15
• Subjects: Artificial neural network -- Dynamic system modelling -- Boltzmann machines -- Diffusion network -- Stochastic differential equation; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2020.107072
• Languages: English
• ISSNs: 0888-3270

Abstract

Graphical abstract: Abstract: The modelling of dynamic system is a challenging problem in a large number of applications like prediction, bio-data modelling, computer vision or time-series processing. To face the complexity and the non-linearity of data, new models are regularly proposed through the literature. Among proposed models artificial neural network (ANN) have benefit of a large interest in the scientist community. The use of latent variables to extract and diffuse complex features in multilayer feedforward neural networks provide usually excellent results. In 1982, Hopfield proposes a generative and deterministic neural network to model a physical system. His work leads to the emergence of a large number of generative neural networks: Boltzmann Machine and its extensions. Different applications lead researchers to propose new extensions for the Boltzmann machine to handle dynamic systems, continuous variables or systems with complex features. In parallel, a new model named the Diffusion Network has emerged, also inspired from Hopfield network but with continuous stochastic properties and designed to solve stochastic differential equations.