Feed-forward neural network unfolding

Abstract

A feed-forward neural network is demonstrated to efficiently unfold the energy distribution of protons and alpha particles passing through passive material. This model-independent approach works with unbinned data and does not require regularization. The training dataset was produced with the same Monte Carlo simulation framework used by the AlCap experiment. The common problem of designing a network is also addressed by performing a hyperparameter space scan to find the best network geometry possible within reasonable computation time. Finally, a comparison with other unfolding methods such as the iterative d'Agostini Bayesian unfolding, and Singular Value Decomposition (SVD) are shown.