Visualization of one NN from the committee predicting a shear point estimate ĝ₁, trained for the fiducial experiment. The NN nodes are represented by black squares, in a configuration with only two hidden layers of five nodes each. The connections between these nodes depict the values of the NN weights, by their thickness and color. The NN biases are visualized by the triangles below the nodes. The legend gives the scale of these elements. From the relative amplitude of the weights, one can observe, for example, that adamom_g2 has relatively low impact, while adamom_sigma plays an important role in the prediciton of ĝ₁. All input features are normalized prior to entering the NNs (see Sect. 3.4 for details).