Abstract:Predicting the evolution of systems that exhibit spatio-temporal dynamics in response to external stimuli is a key enabling technology fostering scientific innovation. Traditional equations-based approaches leverage first principles to yield predictions through the numerical approximation of high-dimensional systems of differential equations, thus calling for large-scale parallel computing platforms and requiring large computational costs. Data-driven approaches, instead, enable the description of systems evolution in low-dimensional latent spaces, by leveraging dimensionality reduction and deep learning algorithms. We propose a novel architecture, named Latent Dynamics Network (LDNet), which is able to discover low-dimensional intrinsic dynamics of possibly non-Markovian dynamical systems, thus predicting the time evolution of space-dependent fields in response to external inputs. Unlike popular approaches, in which the latent representation of the solution manifold is learned by means of auto-encoders that map a high-dimensional discretization of the system state into itself, LDNets automatically discover a low-dimensional manifold while learning the latent dynamics, without ever operating in the high-dimensional space. Furthermore, LDNets are meshless algorithms that do not reconstruct the output on a predetermined grid of points, but rather at any point of the domain, thus enabling weight-sharing across query-points. These features make LDNets lightweight and easy-to-train, with excellent accuracy and generalization properties, even in time-extrapolation regimes. We validate our method on several test cases and we show that, for a challenging highly-nonlinear problem, LDNets outperform state-of-the-art methods in terms of accuracy (normalized error 5 times smaller), by employing a dramatically smaller number of trainable parameters (more than 10 times fewer).

Latent Dynamic Networked System Identification with High-Dimensional Networked Data

Identifiability and Identification of Switching Dynamical Networks: A Data-Based Approach

Identifying Latent Dynamic Components in Biological Systems

Dynamic Processes Modeling and Monitoring Based on a Novel Dynamic Latent Variable Model.

Weighted Linear Dynamic System for Feature Representation and Soft Sensor Application in Nonlinear Dynamic Industrial Processes.

Latent Dynamics Networks (LDNets): learning the intrinsic dynamics of spatio-temporal processes

Dynamic Bayesian Network for Robust Latent Variable Modeling and Fault Classification

Dynamic Probabilistic Latent Variable Model for Process Data Modeling and Regression Application

Probabilistic Sequential Network for Deep Learning of Complex Process Data and Soft Sensor Application

System-Identification Using A New Dynamic Neural-Network

Deep Learning of Latent Variable Models for Industrial Process Monitoring

Network Identification with Latent Nodes Via Autoregressive Models.

Weighted Nonlinear Dynamic System for Deep Extraction of Nonlinear Dynamic Latent Variables and Industrial Application

Learning the intrinsic dynamics of spatio-temporal processes through Latent Dynamics Networks

Direct System Identification of Dynamical Networks with Partial Measurements: a Maximum Likelihood Approach

Bayesian Network for Dynamic Variable Structure Learning and Transfer Modeling of Probabilistic Soft Sensor

Autoregressive Dynamic Latent Variable Models for Process Monitoring

Identification of Linear Networks with Latent Nodes.

On Identification of Dynamical Structure Functions: A Sparse Bayesian Learning Approach.

Unraveling Low-Dimensional Network Dynamics: A Fusion of Sparse Identification and Proper Orthogonal Decomposition

Decentralized System Identification Method for Large-Scale Networks