Accuracy of the trained DNN for the power spectrum convolution. Top left: function R_P(k) obtained with the convolution integral in Eq. (1) (orange triangles) compared with the DNN model (black dots) for one test sample. Top right: relative error of the DNN model in the same test sample used in the left panel. Bottom left: MAEs for all the test samples (the one used in the top panels is highlighted in red and surrounded by a square). Bottom right: 50th, 68th, and 95th error percentiles of the DNN model as a function of k. Generally, the DNN model yields sub-per cent accuracy.