The main idea is to combine "fuzzy logic" with "neural networks". The goal is to construct a supervised learning algorithm based on human understandable rules. The rules are typically (but not constrained to) combinations of AND, OR and NOT operations applied to the features. The neural network is used for finding the best rules. The fuzzy part has to do with assigning continuous values of the input features into categories like "very low", "low", "medium", "high" and "very high".

Need access to the slides...