Fig. 3: Single neuron minimodels achieve similar performance to the 16-320 model. | Nature Communications

Fig. 3: Single neuron minimodels achieve similar performance to the 16-320 model.

From: A simplified minimodel of visual cortical neurons

Fig. 3

a, b Performance of the 16-320 model does not improve with more neurons but improves with more training images. Error bars represent s.e.m. c Minimodel architecture consisting of a fixed conv1 layer from the 16-320 model, and separate conv2 and readout weights for each neuron. d Prediction performance of the minimodel compared to the 16-320 model for each neuron. e Same as (d) summarized per mouse/monkey. Error bars represent s.e.m. f Distribution of the number of conv2 feature maps in the mini-models.

Back to article page