Using Generative Adversarial Networks to biaxially unfold Beta-Oslo Matrices

We explore the use of Generative Adversarial Networks (GANs) to biaxially unfold Beta-Oslo matrices to compute neutron capture cross sections. The Beta-Oslo method, for which the matrices are named, is a technique used to simultaneously extract the Nuclear Level Density and gamma-strength function of neutron-rich nuclei. These properties are important to the calculation of the neutron capture cross section, a nuclear parameter with particular relevance to r-Process nucleosynthesis reaction network calculations. However, the current method for unfolding Beta-Oslo matrices is uniaxially restricted, constraining the amount of information available for analysis. GANs, a state of the art generative modeling technique based on recent advances in deep learning, approach biaxial unfolding as an image-to-image translation problem. Preliminary results from training a Pix2Pix GAN architecture with Beta-Oslo matrices simulated based on the response of the SuN total absorption spectrometer indicate that gamma ray energies can be extracted within a resolution of 6%.