(Caption: a) Examples of equivalent spin models for systems with 6 binary variables; b) High order structure of the voting patterns of the US supreme court during the second Rehnquist Court (1994-2005))

It is common to analyze high dimensional data by projecting the data onto pairwise patterns. Using spin models, we develop new tools to uncover the hidden structures of high dimensional binary data beyond pairwise. Our approach is based on Bayesian model selection and information theory, and puts an emphasis on finding the simplest explanation for the data. Examples of datasets that can be investigated with the techniques we develop include neuronal activity, voting data, stock market data, data for medical diagnosis, animal behavioral data.

Publication highlights: