Fig. 4
From: Brain-guided convolutional neural networks reveal task-specific representations in scene processing
Illustration of the network training and evaluation pipeline for a single image. The network was trained stochastically. Each time an image was passed through the network, it was accompanied by either neural data from the object task or the function task. Once the network was able to classify the task at ~ 90% accuracy, we computed activation maps for each layer and each task.
