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).

Revealing hidden dynamics from time-series data by ODENet

Beyond Predictions in Neural ODEs: Identification and Interventions.

KAN-ODEs: Kolmogorov-Arnold Network Ordinary Differential Equations for Learning Dynamical Systems and Hidden Physics

Time-Reversal Symmetric ODE Network

On the balance between the training time and interpretability of neural ODE for time series modelling

Homotopy-based training of NeuralODEs for accurate dynamics discovery

ODEN: A Framework to Solve Ordinary Differential Equations using Artificial Neural Networks

Universal approximation property of ODENet and ResNet with a single activation function

Latent ODEs for Irregularly-Sampled Time Series

Invariant Neural Ordinary Differential Equations

Neural ODEs with Irregular and Noisy Data

Calibrating multi-dimensional complex ODE from noisy data via deep neural networks

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

Latent Space Energy-based Neural ODEs

Capturing Actionable Dynamics with Structured Latent Ordinary Differential Equations

A numerical method to solve PDE through PINN based on ODENet

NODER: Image Sequence Regression Based on Neural Ordinary Differential Equations

Time Dependence in Non-Autonomous Neural ODEs

RODE-Net: Learning Ordinary Differential Equations with Randomness from Data

Explainable Tensorized Neural Ordinary Differential Equations for Arbitrary-step Time Series Prediction

Explainable Tensorized Neural Ordinary Differential Equations forArbitrary-step Time Series Prediction