Honours Thesis Presentation - Emerging Complexity: Genetic Programming Approaches for Image Recognition
Title: Emerging Complexity: Genetic Programming Approaches for Image Recognition
Supervisor: Dr. Malcolm Heywood
Reader: Dr. Andrew McIntyre
Abstract:
Genetic programming is a branch of machine learning that stochiastically explores a problem's solution space by evolving candidate solutions. In contrast with deep learning, this aims to incrementally build complexity in models.
This work compares several genetic programming models on an image classification task commonly used in deep learning. These approaches are intended to be generic, as opposed to previous models designed to be specific to image recognition. By using this embedded algorithm approach, feature importance is learned alongside model building. This is compared with other incremental approaches to analyze performance and comparative complexity outside of the realm of deep learning algorithms. It is concluded that the generic genetic programming methods reviewed do not reach the level of classification performance provided by deep learning, but do exhibit significantly less complexity.
Time
Location
Room 211, Goldberg Computer Science Building