Applications of Neural Network in Physics: Cosmology and Molecular Dynamics

Understanding the underlying physics of a physical system at both the cosmological and molecular scales has been a focus of attention for decades. Modeling the system using ordinary and partial differential equations along with the Markov Chain Monte Carlo technique are the conventional methods being used. These methods have been proven to reconcile accurate results, however, they fail miserably when the physics is not completely known, which leads to the presence of a large number of free parameters in the model describing the system. Recently, conventional methods have been aided by the use of machine learning techniques to solve real-world problems, which include the use of artificial neural networks such as convolutional neural networks, generative adversarial networks, and random forests. The ability of these techniques to understand the complexity of a physical system and predict new physics solely from data has given a new edge to conventional methods. Their prevalent applications lie in parameter prediction, where available data is used to train a neural network model, and then physical quantities are predicted using the trained model. Classification is another fundamental aspect of machine learning that involves predicting the specific family or category to which the provided data pertains. These techniques find an essential place in physics, providing important insights into complex systems.

Keywords: Non-linear dynamics, Recurrent neural network, Reservoir computing, Transients.