White Learning: A White-Box Data Fusion Machine Learning Framework for Extreme and Fast Automated Cancer Diagnosis

A framework of white learning is proposed in this article, which embraces three categories of white learning models where various levels of hybridization of Bayesian networks and neural networks are fused. At the algorithm level, Bayesian networks and neural computing are integrated tightly as a whole or partial redesigned entity of computing logics. Elements of Bayesian networks and neural computing co-exist in the design of program codes. This level of integration often requires a high extent of intellectual innovation, especially if the new hybrid after coupling the white-andblack box learning models would outperform either one of the original twomodels. On the other hand, loosely coupledmodels are those which run almost independently of each other; exemplars are those ensembles from which the results of the best performing model out of many are taken as the final results. These models are often taken as they are, without any modification in their codes.