Improving classification on images by extracting and transferring knowledge in genetic programming

Abstract

Genetic programming (GP) is a well established evolutionary computation technique that automatically generates a computer program to solve a given problem. GP has been successfully used to solve optimization, symbolic regression and classification problems. Transfer learning in GP has been investigated to learn various Boolean and symbolic regression problems. However, there has been not much work on transfer learning in GP for image classification problems. In this paper, we propose a new technique to use transfer learning in GP to learn image classification problems. The developed method has been compared with the baseline GP method on three image classification benchmarks. The obtained results indicate that transfer learning has significantly improved the classification accuracy in learning various rotated and noisy versions of the tested image classification problems.