Figure 5
From: An adaptive approach to machine learning for compact particle accelerators
(A) The measured beam output is fed into the CNN of convolutional layers with LeakyReLU activation functions followed by dense layers with ReLU activations. The CNN outputs principal component coefficients \((\alpha _1,\dots ,\alpha _N)\) which generate an input beam distribution (B) used as an input to the simulation (C). (D) Simulation output is compared to the measured beam output to adaptively fine tune both the principal components (E) and model parameters (F).
