What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to address the challenge of estimating epidemiological parameters based on partially observed data. Specifically, the paper focuses on how to use Physics - Informed Neural Networks (PINNs) to accurately estimate the unknown parameters in epidemiological models and predict unobserved variables when only partial observed data are available and there is noise. #### Research Background In practical applications, epidemiological models (such as SIR and SEIR models) are used to describe the state changes of different populations during the spread of infectious diseases. These models rely on a set of parameters, such as the infection rate, the latent - period conversion rate, and the removal rate. However, in reality, we often cannot obtain the complete trajectory data of all population states, and the existing observed data may contain noise. This makes it very difficult to estimate these parameters using traditional methods. #### Main Problems 1. **Partially Observed Data**: Not all trajectory data related to the population can be obtained, and some observed trajectories also have noise. 2. **Challenges in Parameter Estimation**: When only partial observed data are available, it is very challenging to use traditional PINNs for parameter estimation. #### Solutions To solve the above problems, the author introduced the concept of algebraic observability and applied it to PINNs. Through this method, even when only partial observed data are available, the unknown parameters can be estimated more accurately and the unobserved variables can be predicted. #### Specific Methods - **SEIR Model**: As an example, the paper selected the SEIR model to describe the spread of infectious diseases: \[ \begin{aligned} \dot{S} &= -\beta SI, \\ \dot{E} &= \beta SI - \epsilon E, \\ \dot{I} &= \epsilon E - \gamma I, \\ \dot{R} &= \gamma I, \\ S + E + I + R &= 1. \end{aligned} \] where \( S \), \( E \), \( I \) and \( R \) represent the proportions of susceptibles, exposed, infected and removed respectively; \( \beta \), \( \epsilon \) and \( \gamma \) represent the infection rate, the latent - period conversion rate and the removal rate respectively. - **Algebraic Observability**: By introducing algebraic observability, polynomial equations can be constructed to represent the relationships between unobserved variables, observed variables and their derivatives. For example, for \( E \) in the SEIR model, its algebraic observability can be represented by the following equation: \[ E=\frac{\dot{I}+\gamma I}{\epsilon}. \] - **Improved PINNs**: Based on algebraic observability, Algebraically Observable PINNs were proposed. Through this improved PINNs, data of unobserved variables can be generated when only partial observed data are available, and parameter estimation can be further optimized. #### Experimental Results The effectiveness of the proposed method was verified through numerical experiments. The experimental results show that, compared with the baseline method (i.e., traditional PINNs), Algebraically Observable PINNs can estimate unknown parameters more accurately and better fit the observed data and predict unobserved variables. #### Summary This paper solves the problem of estimating epidemiological parameters under the condition of partially observed data by introducing the concept of algebraic observability, and improves the accuracy of parameter estimation and the model prediction ability.

Estimate Epidemiological Parameters given Partial Observations based on Algebraically Observable PINNs

A Physics-Informed Neural Network approach for compartmental epidemiological models

On the estimation rate of Bayesian PINN for inverse problems

Nonlinear Discrete-Time Observers with Physics-Informed Neural Networks

Improving PINNs By Algebraic Inclusion of Boundary and Initial Conditions

PINN training using biobjective optimization: The trade-off between data loss and residual loss

Euler iteration augmented physics-informed neural networks for time-varying parameter estimation of the epidemic compartmental model

Randomized Physics-Informed Neural Networks for Bayesian Data Assimilation

Error-Aware B-PINNs: Improving Uncertainty Quantification in Bayesian Physics-Informed Neural Networks

CP-PINNs: Data-Driven Changepoints Detection in PDEs Using Online Optimized Physics-Informed Neural Networks

B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data

A PINN approach for the online identification and control of unknown PDEs

Approaching epidemiological dynamics of COVID-19 with physics-informed neural networks

Convergence and error analysis of PINNs

Estimates on the generalization error of Physics Informed Neural Networks (PINNs) for approximating a class of inverse problems for PDEs

Response Estimation and System Identification of Dynamical Systems via Physics-Informed Neural Networks

A PINN Approach to Symbolic Differential Operator Discovery with Sparse Data

Physics-Informed Neural Networks Integrating Compartmental Model for Analyzing COVID-19 Transmission Dynamics.

Conformalized Physics-Informed Neural Networks

Physics-informed Neural Networks with Unknown Measurement Noise

iPINNs: incremental learning for Physics-informed neural networks