Using Gaussian distribution to construct fitness functions in genetic programming for multiclass object classification

Abstract

This paper describes a new approach to the use of Gaussian distribution in genetic programming (GP) for multiclass object classification problems. Instead of using predefined multiple thresholds to form different regions in the program output space for different classes, this approach uses probabilities of different classes, derived from Gaussian distributions, to construct the fitness function for classification. Two fitness measures, overlap area and weighted distribution distance, have been developed. Rather than using the best evolved program in a population, this approach uses multiple programs and a voting strategy to perform classification. The approach is examined on three multiclass object classification problems of increasing difficulty and compared with a basic GP approach. The results suggest that the new approach is more effective and more efficient than the basic GP approach. Although developed for object classification, this approach is expected to be able to be applied to other classification problems.Keywords Probability based genetic programming, object recognition, object detection, fitness function, multiclass classification Author InformationThe first author is a staff member in computer science, Victoria University of Wellington, New Zealand. The second author is postgraduate student in computer science, Victoria University of Wellington, New Zealand. Using Gaussian Distribution to Construct Fitness Functions in Genetic Programming AbstractThis paper describes a new approach to the use of Gaussian distribution in genetic programming (GP) for multiclass object classification problems. Instead of using predefined multiple thresholds to form different regions in the program output space for different classes, this approach uses probabilities of different classes, derived from Gaussian distributions, to construct the fitness function for classification. Two fitness measures, overlap area and weighted distribution distance, have been developed. Rather than using the best evolved program in a population, this approach uses multiple programs and a voting strategy to perform classification. The approach is examined on three multiclass object classification problems of increasing difficulty and compared with a basic GP approach. The results suggest that the new approach is more effective and more efficient than the basic GP approach. Although developed for object classification, this approach is expected to be able to be applied to other classification problems.

Searching for Novel Classifiers

“…This work uses the Static Range Selection GP Classifier (SRS-GPC) described by Zhang and Smart [17]. In a classification problem, a pattern x ∈ R P has to be classified as belonging to a single class from Ω = {ω 1 , ..., ω M }, where each ω i represents a distinct class label.…”

Searching for novel regression functions

“…For example, consider the GP classifier based on static range selection (SRS) [35], that functions as follows: for a two class problem and real-valued GP outputs, the SRS classifier is straightforward; if the program output for input pattern x is greater than zero then the pattern is labeled as belonging to class A, otherwise it is labeled as a class B pattern. In this case, while the semantic space description (as defined above) of two programs might be different (maybe substantially), they can still produce the same high-level classification (consider any two outputs y 1 , y 2 ∈ (0, ∞) with y 1 = y 2 ).…”

Preliminary Study of Bloat in Genetic Programming with Behavior-Based Search

“…However, strictly focusing on program outputs might not be the best approach in some domains. For example, consider the GP classifier based on static range selection (SRS) [38] (it will be further discussed in Section 4 and used in the experimental work), that functions as follows. For a two class problem and realvalued GP outputs, the SRS classifier is straightforward; if the program output for input pattern x is greater than zero then the pattern is labeled as belonging to class A, otherwise it is labeled as a class B pattern.…”

“…This work uses the Static Range Selection GP Classifier (SRS) described by Zhang and Smart [38]. In a classification problem, a pattern x ∈ R p has to be classified as belonging to a single class from Ω = {ω 1 , ..., ω M }, where each ω i represents a distinct class label.…”