 
Summary: Frontiers in Computational Neuroscience www.frontiersin.org August 2009  Volume 3  Article 11  1
COMPUTATIONAL NEUROSCIENCE
ORIGINAL RESEARCH ARTICLE
published: xx August 2009
doi: 10.3389/neuro.10.011.2009
et al., 2009). We found that useful learning is always still possible
provided that Hebbian adjustments retain some connection spe
cificity, though it is degraded. However, there has been increasing
realization that at least in the neocortex unsupervised learning
must be sensitive to higherthanpairwise correlations, which
requires that the learning rule at individual connections be non
linear. Since the number of possible higherorder correlations is,
for highdimensional input patterns, essentially unlimited, useful
learningmightrequirethattheconnectionlevelnonlinearlearning
rule be extremely accurate.
To test this idea, we studied the effect of introducing Hebbian
crosstalk in perhaps the simplest neural network model of non
linear learning, independent components analysis (ICA) (Bell and
Sejnowski, 1995; Hoyer and Hyvärinen, 2000; Hyvärinen et al.,
2001; Nadal and Parga, 1994). In this model, it is assumed that the
