Abstract
This project examines and develops a powerful technique for producing approximate solutions to partial differential equations using machine learning. We expand on the existing literature and analyze the effects of various parameters used in the employed model architecture, including the number of layers, neurons, epochs, batches, the activation function, and the optimization method. It appears that techniques based on Neural Networks potentially pave a new route to construct provably efficient algorithms for solving high-dimensional and nonlinear PDEs. We provide a physics-informed neural network model to approximate a specific category of the parabolic differential equations famously known as Burger's equation. We designed a calculator-like user interface that asks for user input for various equation boundaries to employ the physics-informed neural network to approximate solutions and display various parameters on the user interface for ease of usability and visibility.