Fig. 2: Number of convolutional feature maps required to fit visual responses and perform visual tasks. | Nature Communications

Fig. 2: Number of convolutional feature maps required to fit visual responses and perform visual tasks.

From: A simplified minimodel of visual cortical neurons

Fig. 2

a Schematic of the simplified two-layer model. b Performance of the model as a function of the number of convolutional channels in the first layer (conv1) and the second layer (conv2) across 6 mice (top) and 2 monkeys (bottom). c Same data as (b) displayed as curves. d Conv1 weights for mouse (top) and monkey (bottom). e Two-layer convolutional models were trained to perform image classification on the ImageNet dataset or to perform texture classification on a dataset with 201 natural texture images. f, g Model performance as a function of the number of channels in conv1 and conv2 for texture classification and ImageNet classification.

Back to article page