Physics-Informed Neural Networks (PINNs) employ neural networks to solve supervised learning tasks while respecting underlying physical laws expressed as partial differential equations. The framework provides two main algorithmic approaches: continuous-time and discrete-time models depending on data availability. It enables both inferring solutions to PDEs, yielding differentiable surrogate models, and discovering unknown PDEs from observational data. This approach creates data-efficient universal function approximators that embed physical laws as prior information.