Optimizing multilayer Bayesian neural networks for evaluation of fission yields

The Bayesian machine learning is a promising tool for the evaluation of nuclear fission data but its potential capability has not been fully realized. We attempt to optimize the performances of the multilayer Bayesian neural networks for evaluations of fission yields. The influences of adjustments of learning data, activation functions, network structures have been studied. In particular, negative values of net functions have been penalized to avoid non-physical inferences of fission yields. Presently the network with double hidden layers has optimal performances compared to the singlelayer or deeper networks. These studies are essential for further developments of precise evaluation methods.