INNE: a structured learning algorithm for noisy examples
Michel Liquière, Jean Sallantin · 2003
A technique for learning structural concepts from noisy examples is presented. Using the description language defined by J.F. Sowa (1984) provides a convenient way of expressing the knowledge and the properties used by the algorithm. This graph description makes it possible to better manage learning problems, define new methods, and present results in a familiar and practical way. A learning system INNE has been developed which is based on this kind of description language. The goal is to design a learning algorithm which allows the processing of a considerable amount of data in a reasonable time. This kind of learning engine has been successfully tested on a real problem in biology. Thus, an experimental basis and a validation of the method have been acquired.>