DynoNet: A neural network architecture for learning dynamical systems. (14th January 2021)

• Record Type: Journal Article
• Authors: Forgione, Marco; Piga, Dario
• Is Part Of: International journal of adaptive control and signal processing. Volume 35:Number 4(2021)
• Page Start: 612
• Page End: 626
• Publication Date: 2021-01-14
• Subjects: machine learning -- neural networks -- system identification; Adaptive control systems -- Periodicals; Adaptive signal processing -- Periodicals; 629.836
• DOI: 10.1002/acs.3216
• Languages: English
• ISSNs: 0890-6327

Abstract

Summary: This article introduces a network architecture, called dynoNet, utilizing linear dynamical operators as elementary building blocks. Owing to the dynamical nature of these blocks, dynoNet networks are tailored for sequence modeling and system identification purposes. The back‐propagation behavior of the linear dynamical operator with respect to both its parameters and its input sequence is defined. This enables end‐to‐end training of structured networks containing linear dynamical operators and other differentiable units, exploiting existing deep learning software. Examples show the effectiveness of the proposed approach on well‐known system identification benchmarks.