Measuring Information Transfer

CSSS Santa Fe 2007 Home Faculty U.S. Visa Requirements Participants Schedule Readings Logistics Travel Projects & Working Groups Tutorials After Hours Photos T-shirt Design Forum Project Presentations Final Papers

Background

Dependencies between variables or processes in systems are often quantified using correlation coefficients (income <-> height, etc.). However, correlation only captures linear dependencies. In addition the dynamical aspect (e.g. common history) and the direction of information transfer in the processes under consideration are neglected. A recently proposed measure called Transfer Entropy was specifically designed to overcome those limitations by the clever use of conditioned transition probabilities.

In theory, Transfer Entropy, is able to quantify non-/linear directed dependencies in any dynamical system. In practice, the transition probabilities/-densities have to be estimated from the data.

That's the reason, why we need your data! If you happen to have good multivariate time series data (e.g. bio-chemical/medical/logical dynamical systems, longitudinal studies in sociology, etc.), which you expect to be holding interesting dependencies of any kind, let us know.

Team members (feel free to join)

• Juergen_Pahle
• Joseph_Lizier

Readings

• T. Schreiber (2000) Measuring information transfer. Phys. Rev. Lett 85(2), 461-4

Download at [[1]] (requires access to PRL).

Questions to answer

• What can we learn about (possibly nonlinear) dependencies between variables in biomedical, ecological, sociological ... systems
• How does information transfer measured by the transfer entropy breaks down in systems, on which damage was inflicted (ischemia in heart, extinction of species in dynamical foodwebs etc.)
• ...

Ideas

• Measure information transfer in the heart model by Vikas, Simon, Nathan & Co.
• ...