Fast Algorithms for the Combination and Creation of Multiple Classifiers using Subsampling

• Grant number: 11/01282-4
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: May 01, 2011
• End date: April 30, 2013
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Moacir Antonelli Ponti
• Grantee: Isadora Rossi
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Multiple classifier combination methods have been demonstrated to be effective, under some conditions, for several pattern recognition applications. These methods are often used to improve performance of instable or weak classifiers. The studies in this field do not, in general, address aspects related to the speed of the methods. In applications with large data sets the sampling is a challenge, since training and classification tasks could become slow when using common rates of sampling. With this in mind, the investigation of fast methods to create and combine multiple classifiers is important to several applications. In this project, fast classifier combination methods are going to be studied and developed. Questions such as generation of classifier ensembles, data subsampling and efficient combination will be addressed. The relation between performance and speed is also an important point in this study. Experiments with simulated and real data set will be carried out, with an additional assessment through a visualization tool for classification tasks.