Incremental supervised learning in optimum-path forests

• Grant number: 14/04889-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: June 01, 2014
• End date: July 31, 2016
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Rodrigo Fernandes de Mello
• Grantee: Mateus Riva
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated scholarship(s): 15/24652-2 - Anomaly detection using an incremental learning algorithm based on minimum spanning tree, BE.EP.IC

Abstract

The incremental learning approach needs the dynamic addition of information extracted from observations. This kind of learning is important in applications in which the observations are not available for the creation of the classifier since the beginning of the process. Thus, multiple training procedures are necessary to complete the learning. The optimum-path forest classifier (OPF) is currently been used in several applications. However, the original version does not include the incremental learning feature. Therefore, in this project we expect to contribute with the development of an algorithm capable of including new nodes on the optimum path trees, reducing the running time of the training process after the inclusion of new labeled examples. Also, we will investigate the removal of trees, reconstructing the model so that it better fits the current training sample.