Abstract
This disclosure describes a neuro-fuzzy controller that can be utilized to guide image classifier networks for classification of subjective attributes. Per techniques of this disclosure, linguistic expert rules for memberships of an image to various output categories of the subjective attribute(s) are framed and the classification is analyzed as a fuzzy system. Fuzzy rules and fuzzy inference output from this system are used to guide a neural network to effectively incorporate the expert rules. Specific loss functions are utilized to guide the image classifier. A fuzzy-rule contradiction loss is utilized to capture a weighted deviation of image classifier prediction from expert rules. A fuzzy inference loss is utilized to capture overall deviation from fuzzy inference output. Utilization of the neuro-fuzzy controller can enable image classifier models to classify images according to subjective attributes, e.g., to provide accurate labels for family friendliness of a restaurant based on images of the restaurant.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Max, Lenord Melvix Joseph Stephen, "Guiding Image Classifier Using a Neuro-fuzzy Controller", Technical Disclosure Commons, (October 30, 2023)
https://www.tdcommons.org/dpubs_series/6371