Methods of Pattern Inference from Symbol Sequences: Practical Computational Mechanics
Nonlinear Time Series Analysis Seminar
Max Planck Institute for the Physics of Complex Systems
February 2005
Abstract:
I will discuss several algorithms for pattern inference, or
epsilon-machine reconstruction, from data. If the data come in the form of
a symbol sequence, there are two standard inference algorithms available:
the subtree merging method and causal state splitting reconstruction. If
instead, the data are power spectra, other techniques are used, such as
epsilon-machine spectral reconstruction or reverse Monte Carlo methods. I
will contrast these methods with several examples.