Unsupervised clustering grouped the pictures in groups similar to the ones used for decoding

Results of unsupervised clustering of the neuronal similarity matrix. We defined the neuronal similarity between two pictures i and j based on the spiking population activity as the Pearson correlation coefficient between the vectors containing responses of all neurons (n=367) to picture i and picture j (see Figure with neuronal similarities). The responses consisted of the spike count in the [100;300] ms interval after stimulus onset. The dimensions of the resulting similarity matrix were 77 x 77 (symmetric matrix). We performed a k-means clustering algorithm (with 10 iterations and random initial conditions) on the neuronal similarity matrix. The rectangles delimit the pictures belonging to the same cluster.

Results with 2 clusters
Results with 3 clusters

Results with 4 clusters
Results with 5 clusters
Results with 6 clusters
Results with 7 clusters
Results with 8 clusters
Results with 9 clusters
Results with 10 clusters

Top