Stochastic processes with memory: Augmented toolbox for data analysis
Abstract
Rapid increase in computational power has allowed extraction of valuable insights on our environment aided by visualization of large datasets in terms of connectivity, graphs, histograms, and time series, among others. Analytical modelling, however, could further provide deeper understanding of various data-rich natural and social phenomena.
In this talk, we show that a number of complex systems generating seemingly random data x can exhibit interesting patterns when mathematically modeled with time evolution as [1],
x(T) = x0 + g(T) ∫0T f(T-t) h(t) dB(t),
where x0 is the initial value and B(t) $ is the Wiener process. Here, the memory function f(T-t) and time-dependent functions h(t) and g(t) are chosen depending on the phenomenon being investigated. This describes a large class of stochastic processes with memory for which the probability density function is obtained exactly. We give as illustrative examples stochastic analysis for large datasets arising from geomagnetic fluctuations, stock market price fluctuations, and diffusion of proteins [2].
[1] C. C. Bernido and M. V. Carpio-Bernido, Methods and Applications of White Noise Analysis in Interdisciplinary Sciences (World Scientific, Singapore, 2014).
[2] W. I. Barredo, C. C. Bernido, M. V. Carpio-Bernido, M. G. O. Escobido, P. J. Ong, R. R. Violanda, A. Yoshikawa, T. Uozumi, S. Abe, and A. Fujimoto, Group Technical Report (2016).