What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to explore and compare the applications of two different generator sub - forms (single generator and double generator formalisms) in Thermodynamics - Informed Neural Networks (TINNs). Specifically, the paper focuses on: 1. **Introducing physical principles to improve prediction accuracy**: - The inductive biases of neural networks have been proven to significantly improve the accuracy and robustness of their predictions of physical phenomena. These biases not only reduce prediction errors but also allow training with smaller data sets. 2. **Ensuring the consistency of thermodynamic principles**: - When there is a lack of knowledge of the specific physical principles of the studied phenomena, thermodynamic principles provide a generally valid constraint. Therefore, the paper focuses on how to ensure that the learning process of the neural network complies with these basic physical laws by introducing thermodynamic principles. 3. **Analysis of the advantages and disadvantages of single - generator and double - generator forms**: - In the dynamic description of non - equilibrium systems, single - generator and double - generator forms are two commonly used methods. Through the study of classical problems and synthetic data, the paper analyzes the advantages and limitations of these two forms, especially their influence on maintaining the system structure during the learning process. 4. **Performance evaluation in practical applications**: - The paper evaluates in detail the performance of these two generator forms in practical applications through two specific problems - discrete problems (such as double - thermoelastic pendulum) and continuous problems (such as Couette flow of Oldroyd - B fluid), and draws conclusions about their respective advantages and disadvantages. ### Summary of mathematical formulas - **Generalized free energy**: \[ F = H+S \] where \(H\) is the Hamiltonian and \(S\) is the entropy. - **Equations of motion**: For the single - generator form: \[ \frac{dz}{dt}=\{\{z, F\}\} \] For the double - generator form: \[ \frac{dz}{dt}=L\frac{\partial(H + S)}{\partial z}+M\frac{\partial(H + S)}{\partial z} \] - **Energy conservation and entropy increase conditions**: \[ \frac{dH}{dt} = 0 \] \[ \frac{dS}{dt}\geq0 \] Through these formulas, the paper shows how to embed thermodynamic principles into neural networks to ensure that the learning results of the model not only conform to physical laws but also have high prediction accuracy.

A comparison of Single- and Double-generator formalisms for Thermodynamics-Informed Neural Networks

A comparison of single and double generator formalisms for thermodynamics-informed neural networks

Graph neural networks informed locally by thermodynamics

Advancements in Thermochemical Predictions: A Multi-Output Thermodynamics-Informed Neural Network Approach

Improving Data-Driven Inferential Sensor Modeling by Industrial Knowledge: A Bayesian Perspective

A statistical mechanics framework for Bayesian deep neural networks beyond the infinite-width limit

Gibbs-Duhem-Informed Neural Networks for Binary Activity Coefficient Prediction

Thermodynamic Overfitting and Generalization: Energetic Limits on Predictive Complexity

A theory of data variability in Neural Network Bayesian inference

The principles of adaptation in organisms and machines II: Thermodynamics of the Bayesian brain

Thermodynamic Fidelity of Generative Models for Ising System

Thermodynamics-informed super-resolution of scarce temporal dynamics data

Partitioned integrators for thermodynamic parameterization of neural networks

On Inductive Biases That Enable Generalization of Diffusion Transformers

InVAErt networks: A data-driven framework for model synthesis and identifiability analysis

Thermodynamic Bayesian Inference

Learning Feynman integrals from differential equations with neural networks

The twin peaks of learning neural networks

From statistical inference to a differential learning rule for stochastic neural networks

Increasing biases can be more efficient than increasing weights