Machine learning structure preserving brackets for forecasting irreversible processes

Abstract

Forecasting of time-series data requires imposition of inductive biases to obtain predictive extrapolation, and recent works have imposed Hamiltonian/Lagrangian form to preserve structure for systems with reversible dynamics. In this work we present a novel parameterization of dissipative brackets from metriplectic dynamical systems appropriate for learning irreversible dynamics with unknown a priori model form. The process learns generalized Casimirs for energy and entropy guaranteed to be conserved and nondecreasing, respectively. Furthermore, for the case of added thermal noise, we guarantee exact preservation of a fluctuation-dissipation theorem, ensuring thermodynamic consistency. We provide benchmarks for dissipative systems demonstrating learned dynamics are more robust and generalize better than either "black-box" or penalty-based approaches.

References in corpus (12)

• Adam: A Method for Stochastic Optimization
• Time-series learning of latent-space dynamics for reduced-order model closure
• Achieving Conservation of Energy in Neural Network Emulators for Climate Modeling
• ANODE: Unconditionally Accurate Memory-Efficient Gradients for Neural ODEs
• Hamiltonian Graph Networks with ODE Integrators
• Symplectic Neural Networks in Taylor Series Form for Hamiltonian Systems
• ANODEV2: A Coupled Neural ODE Evolution Framework
• Turbulence forecasting via Neural ODE
• MALI: A memory efficient and reverse accurate integrator for Neural ODEs
• GENERIC: Review of successful applications and a challenge for the future
• Enforcing exact physics in scientific machine learning: a data-driven exterior calculus on graphs
• Coarse-graining via the fluctuation-dissipation theorem and large-deviation theory

Cited by in corpus (4)

• GFINNs: GENERIC Formalism Informed Neural Networks for Deterministic and Stochastic Dynamical Systems
• Structure-preserving Sparse Identification of Nonlinear Dynamics for Data-driven Modeling
• One-Shot Transfer Learning of Physics-Informed Neural Networks
• SlotPi: Physics-informed Object-centric Reasoning Models