Non normalized Redundancy

**Figure:** **S2.** Comparison of un-normalized redundancies in different brain regions. The panels show redundancy calculated using (a) distribution of spike counts (b) spike counts in triplets of cells (c) distribution of first spike latencies (d) the direct method (Spike patterns as binary words). The non-normalized redundancy measures yield larger differences between IC and MGB/A1, presumably because of the higher level of single-unit MI in IC relative to MGB and A1.
$\begin{figure}\centerline{\psfig{silent=,figure=neuron-suppl-2.eps}} \end{figure}$

Normalization of the redundancy measure is required to correct for the effect of single-unit information on pairs redundancy. Different brain areas have different single cell information levels, which impose different limits on the pairwise redundancy in the different brain areas and therefore requires this normalization.

To demonstrate that the normalization does not strengthen the effects we observed but actually weaken it, Fig. S2 compares un-normalized redundancies, which exhibit a much stronger difference between IC and AI/MGB.