Gradient descent training expert system
Jeremy Straub · Software Impacts · 2021
This software is used to build, train and present data for evaluation by a gradient descent training expert system (GDES). A GDES uses a machine learning training method, gradient descent, in a manner similar to a neural network; however, instead of a multi-layer network of densely connected nodes, it uses a known-meaning rule-fact network. Thus, the logical relationships (rules) between nodes (facts) are human (or, potentially in the future, autonomously) defined and can have an identified meaning; however, their weightings are optimized using machine learning techniques. This provides the explainability of an expert system with the optimization of a neural network.