ABSTRACT
This chapter discusses how machine learning can help analyze time-series data, discover governing equations for nonequilibrium systems, or elucidate the systems time-dependent response to environmental cues. It discusses a machine learning-based approach to analyze time-series data and extract the nature of diffusion for a given system. Neural networks can be trained to learn how a system evolves with time and any time-dependent information from time-series data. First, RNNs will allow for the classification of different types of diffusive behavior. Second, RNNs can be used for inference tasks and lead to the prediction of the value for the diffusion exponent. Rather than mapping input and output pairs, the idea is to consider a flow of particles, track how many particles successfully move from the starting point to the endpoint within a given time interval, and use the outcome of the trajectory to define a system of rewards, or “reinforcements” in psychology, to guide the particle towards the target.
