Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

Carleton University

$24,000.00

Machine learning with enhanced evolutionary algorithms

FY 2006-07

Summary
Most Machine Learning (ML) work to date has tackled problems of supervised learning where the data are presented in the form of vectors of 1-place predicates and their values. A supervised learning system attempts to find concept descriptions for classes that are, together with preclassified examples, supplied to it by a teacher. Recently, we have developed and tested two evolutionary ML systems. One of these systems is based on a modification of Genetic Programming (GP) and compares favorably with other supervised learners such as C5 but, unlike the others, also learns when given relational data. The second system is a nearly parameter-free evolutionary algorithm that evolves populations of decision trees. This system significantly outperforms several other supervised learners on commonly used test datasets. We intend to extend our algorithms to enable them to perform unsupervised learning as well. Unsupervised learning tasks cannot rely on a target class and prespecified examples, and range from the discovery of concepts through conceptual clustering to the further discovery of empirical regularities relating concepts constructed by the system. Besides being intrinsically more difficult than supervised learning, unsupervised learning also poses serious problems in the post-analysis of the results because testing the discovered relationships is subject to personal bias. We will develop a testing methodology to address this problem. On the basis of our experience with applying evolutionary algorithms to supervised learning tasks we expect that progress with relational learning from richer representations and with unsupervised learning will require several improvements to the standard GP approach. For this part of the research we intend to strenghten GP by introducing a divide-and-conquer approach to problems, i.e. to investigate how building blocks can be automatically composed into useful higher-level modules.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
313-2006
Name nom
Oppacher, Franz
Department département
Computer Science, School of
Organization id
23
Institution établissement
Carleton University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2006
Competition year année de concours
2006
Award amount
24000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
331
Committee name en
Computing and Information Sciences - B
Committee name fr
Informatique - B
Area of application code
802
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Information systems and technology
Area of application fr
Systèmes d'information et technologie de l'information
Research subject code
2800
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Artificial intelligence (computer vision, use2603)
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
Partie
1
Nb partie
5
Application title
Machine learning with enhanced evolutionary algorithms
Keyword
NA
Application summary
Most Machine Learning (ML) work to date has tackled problems of supervised learning where the data are presented in the form of vectors of 1-place predicates and their values. A supervised learning system attempts to find concept descriptions for classes that are, together with preclassified examples, supplied to it by a teacher. Recently, we have developed and tested two evolutionary ML systems. One of these systems is based on a modification of Genetic Programming (GP) and compares favorably with other supervised learners such as C5 but, unlike the others, also learns when given relational data. The second system is a nearly parameter-free evolutionary algorithm that evolves populations of decision trees. This system significantly outperforms several other supervised learners on commonly used test datasets. We intend to extend our algorithms to enable them to perform unsupervised learning as well. Unsupervised learning tasks cannot rely on a target class and prespecified examples, and range from the discovery of concepts through conceptual clustering to the further discovery of empirical regularities relating concepts constructed by the system. Besides being intrinsically more difficult than supervised learning, unsupervised learning also poses serious problems in the post-analysis of the results because testing the discovered relationships is subject to personal bias. We will develop a testing methodology to address this problem. On the basis of our experience with applying evolutionary algorithms to supervised learning tasks we expect that progress with relational learning from richer representations and with unsupervised learning will require several improvements to the standard GP approach. For this part of the research we intend to strenghten GP by introducing a divide-and-conquer approach to problems, i.e. to investigate how building blocks can be automatically composed into useful higher-level modules.