1: \begin{abstract}

2: Partial Differential Equations (PDEs) are ubiquitous in many disciplines of science and engineering and notoriously difficult to solve. 

3: In general, closed-form solutions of PDEs are unavailable and numerical approximation methods are computationally expensive. 

4: The parameters of PDEs are variable in many applications, such as inverse problems, control and optimization, risk assessment, and uncertainty quantification. 

5: In these applications, our goal is to solve parametric PDEs rather than one instance of them.

6: Our proposed approach, called Meta-Auto-Decoder (MAD), treats solving parametric PDEs as a meta-learning problem and utilizes the Auto-Decoder structure in \cite{park2019deepsdf} to deal with different tasks/PDEs.

7: Physics-informed losses induced from the PDE governing equations and boundary conditions is used as the training losses for different tasks.

8: % Putting them together, we propose a novel method, \textit{Meta-Auto-Decoder} (MAD) for solving parametric PDEs in an unsupervised manner. 

9: The goal of MAD is to learn a good model initialization that can generalize across different tasks, and eventually enables the unseen task to be learned faster. 

10: The inspiration of MAD comes from (conjectured) low-dimensional structure of parametric PDE solutions and we explain our approach from the perspective of manifold learning.

11: Finally, we demonstrate the power of MAD though extensive numerical studies, including Burgers' equation, Laplace's equation and time-domain Maxwell's equations. 

12: MAD exhibits faster convergence speed without losing the accuracy compared with other deep learning methods.

13: \end{abstract}

14: