Supervised learning in CINets
Paul L. Bruhn, Jeffrey J. Weinschenk · 2011
Abstract – Continuous Inference Networks (CINets), a form of multilayer fuzzy value networks, allow computation with fuzzy values in concise structures, are capable of universal function approximation, and are readily interpretable through natural language, aiding maintenance, modification, collaboration, and knowledge sharing. However CINets have been reliant on Subject Matter Expertise (SME) and manual tuning to realize optimal performance, limiting their applicability. With ONR support[i], ARL has developed a supervised learning process for CINets, capable of designing a CINet structure, and of optimizing an existing CINet structure. The CINet supervised learning process allows the automated development of data fusion, classification, and pattern recognition structures that are interpretable, modifiable, and concise. Performance of CINets developed with the supervised learning process is compared to that of Artificial Neural Network (ANNs), fuzzy logic rule set, and Bayesian network approaches.