Marks, Robert J., II (Robert Jackson), 1950-2022-01-282022-01-282021-122021-10-19December 2https://hdl.handle.net/2104/11704Neural networks have often been described as black boxes. The prevalence of publicly available neural networks and the application of transfer learning has allowed for the development of systems with minimal understanding of the data distribution. For example, a generic neural network trained to differentiate between kittens and puppies will classify a picture of a kumquat as either a kitten or a puppy, despite the kumquat residing outside the known data distribution. The neural network watchdog is a technique which screens trained classifier and regression machine input candidates to determine the distribution validity, and allows for methods of out-of-distribution removal with minimal performance impact.application/pdfenNeural networks. Machine learning. Watchdog.ThesisWorldwide access2022-01-28