Classifier performance is significantly worse for arbitrary group definition

Comparison of the classifier performance using the 8 groups of pictures defined in the text (see Methods above) versus using 8 arbitrary groups formed by random collections of pictures. The number and size of the random groups was matched to the corresponding values in the normal groups. Error bars show s.d. over 30 iterations using randomly selected sites. The horizontal dashed line shows chance performance (1/8). Classifier parameters: MUA, n=256 sites, [100;300) ms interval, bin size = 50 ms.

Top