Machine Learning with Templates

New methods are presented for the machine recognition and learning of categories, patterns, and knowledge. A probabilistic machine learning algorithm is described that scales favorably to extremely large datasets, avoids local minima problems, and provides fast learning and recognition speeds. Templates may be created using an evolutionary algorithm described here, constructed with other machine learning methods, designed by a human expert or synthesized using a combination of these methods. Each template has a prototype and matching function which can help improve generalization. These methods have applications in bioinformatics, financial data mining, goal-based planners, handwriting recognition, machine vision, natural language processing / understanding, search engines, strategy such as business and games and voice recognition.