The NOISE dataset: Causal Directions in Noisy Environment

Guido Nolte, Fraunhofer FIRST, Germany

This challenge has two parts, a simulation and real data.
Simulation: Data are simulated as superposition of bivariate unidirectional interaction plus additive mixed and non-white noise. The simulations were done with AR-models with uniformly distributed input. The challenge is to estimate the causal direction. For each out of 1000 examples you get +1 point for the correct answer, -10 points for the wrong answer, and 0 points for no answer.
Real Data: These are high quality EEG data for 10 subjects for 19 channels. The data contain a prominent peak at around 10 Hz predominantly in occipital (back) channels. No ground truth is known. A submission must be a single 19x19 matrix corresponding to a causality estimate between all pairs of channels averaged across subjects. Any submission will be visualized and, with the agreement of the authors, put on the net for an open discussion.


Comparison of Granger Causality and Phase Slope Index

Guido Nolte, Andreas Ziehe (Fraunhofer FIRST, Germany) and Nicole Krämer, Klaus-Robert Müller (TU Berlin, Germany)

We recently proposed a new measure, termed Phase Slope Index (PSI), to estimate the causal direction of interactions designed to be robust to instantaneous mixtures of independent sources with arbitrary spectral content. We compared this method to Granger Causality for linear systems containing spatially and temporarily mixed noise and found that, in contrast to PSI, the latter was not able to properly distinguish truly interacting systems from mixed noise. Here, we extent this analysis with respect to two aspects: a) we analyze Granger causality and PSI also for non-mixed noise, and b) we analyze PSI for nonlinear interactions. We found a) that Granger causality, in contrast to PSI, fails also for non-mixed noise if the memory-time of the sender of information is long compared to the transmission time of the information, and b) that PSI, being a linear method, eventually misses nonlinear interactions but is unlikely to give false positive results.

NIPS 2008 workshop on causality