System modeling using composition networks
The authors present a new approach to the modeling of continuous-time systems in discrete time. The reasoning behind this approach is that discrete-time models of continuous-time systems ought to be invertible. Thus, one should investigate group theoretic methods -- using compositions of invertible basis transformations to approximate the (invertible) map of a dynamical system. It turns out that the appropriate framework is Lie algebra theory (the basics of which are reviewed in the paper) and, more specifically, the use of the powerful Baker-Campbell-Hausdorff formula to develop methods of integration for ordinary differential equations using compositions. This leads to the development of architectures (which are called composition networks and recurrent composition networks) for dynamical system prediction and for nonlinear system identification. These techniques incorporate a priori knowledge about the system (grey-box models) and therefore favorably compare with multilayer perceptrons, if a model of the system is available. For the case where no a priori knowledge is available, the authors merge the composition network and the multi-layer perceptron to produce an `` MLP in dynamic space (notion borrowed from the neural networks theory), which is an architecture having the universal approximation property for dynamical systems. For the entire collection see [Zbl 0892.00024].
- Modelling systems by hypergraphs
- scientific article; zbMATH DE number 1928759 (Why is no real title available?)
- scientific article; zbMATH DE number 5201482 (Why is no real title available?)
- scientific article; zbMATH DE number 2230298 (Why is no real title available?)
- Algebraic dynamical systems in machine learning
- \textit{dynoNet}: a neural network architecture for learning dynamical systems
This page was built for publication: System modeling using composition networks
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4400818)
