An application of neural networks in stochastic physics

Synopsis

Machine learning allows physicists to make use of large amounts of data for studying physics models and making predictions on system dynamics. In non-equilibrium statistical mechanics, the Fokker-Planck equation (FPE) is used to describe how time-dependent probability density functions (PDFs) evolve in a stochastic system. However, it has been difficult to apply and predict PDF evolution with higher accuracy in real-world systems. Neural networks, which are a subset of machine learning, have been used to perform time series prediction and modelling in many fields. In recent studies, the neural network approach shows physicists a path that more accurately predicts PDF evolution where FPE parameters can be obtained by training on a set of PDF data. This might lead to significant improvement in many real-world applications, such as studies on DNA bubble dynamics in biophysics and heat flux evolution in meteorology. In this talk, I will discuss the neural network approach for modelling a system that is under random fluctuations.