Classification with Runge-Kutta networks and feature space augmentation

DOI10.3934/JCD.2021018MaRDI QIDQ2063032zbMATH OpenOpenAlexFDO

Authors Elisa Giesecke, Axel Kröner

Publication date 10 January 2022

Published in Journal of Computational Dynamics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2104.02369

zbMATH Keywords

augmented neural ODEsdeep learningpoint and image classificationRunge-Kutta networks

Mathematics Subject Classification ID

Multistep, Runge-Kutta and extrapolation methods for ordinary differential equations (65L06) Artificial neural networks and deep learning (68T07)

Abstract: In this paper we combine an approach based on Runge-Kutta Nets considered in [Benning et al., J. Comput. Dynamics, 9, 2019] and a technique on augmenting the input space in [Dupont et al., NeurIPS, 2019] to obtain network architectures which show a better numerical performance for deep neural networks in point and image classification problems. The approach is illustrated with several examples implemented in PyTorch.

Recommendations

• scientific article; zbMATH DE number 2201211
• Input-to-state stability for system identification with continuous-time Runge–Kutta neural networks
• Linear feature transform and enhancement of classification on deep neural network
• SRKCD: a stabilized Runge-Kutta method for stochastic optimization
• Stable architectures for deep neural networks

Cites work

• A proposal on machine learning via dynamical systems
• Control under constraints for multi-dimensional reaction-diffusion monostable and bistable equations
• Deep learning as optimal control problems: models and numerical methods
• Deep learning: an introduction for applied mathematicians
• Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Runge-Kutta methods in optimal control and the transformed adjoint system
• Structure-preserving deep learning
• Symplectic Runge-Kutta and related methods: Recent results
• Symplectic Runge-Kutta schemes for adjoint equations, automatic differentiation, optimal control, and more

Cited in

(2)

• Linear feature transform and enhancement of classification on deep neural network
• A hybrid Sobolev gradient method for learning NODEs

Describes a project that uses

Uses Software

• torchdiffeq
• ANODEs
• PyTorch
• GitHub

This page was built for publication: Classification with Runge-Kutta networks and feature space augmentation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2063032)