Extended Data Fig. 6: Layers of the feedforward CNN, adapted from 2.

Conv2D refers to a convolutional layer, F to the number of filters, K to the kernel size, S to the strides, U to the number of units in a fully-connected layer. The values of the dropout-rates δ1,δ2, and δ3 were optimised on the passive benchmark and reported in Extended Data Figure 3.