Dimensionality of activity in different layers of a deep convolutional neural network (CNN).

<p><i>A</i>: The CNN comprised different layers (black outline boxes), where each layer comprised a group of units that performed specific operations, such as convolution, pooling, concatenation, and normalization. For each of the red-outlined units, we assessed the dimensionality of the activity of 100 filters. <i>B</i>: Dimensionality and variance of model responses in each layer to movies, computed in the same manner as in <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1005185#pcbi.1005185.g003" target="_blank">Fig 3<i>C</i></a>.