A lateral symmetry approach to Percentage-based Hybrid Pattern (PHP) training

Sheng Uei Guan*, Kiruthika Ramanathan

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review


In this paper, we investigate the application of lateral symmetry to supervised learning using genetic algorithms. The hypothesis is motivated by the presence of symmetry in the animal brain and by research results showing approximately equal task division between the two hemispheres of the brain. In this paper, each training pattern is considered a task. By applying the concept of lateral symmetry, we use global training (a typically right brained activity) to learn half the tasks and local training (a left brained activity) to learn the rest of the tasks. We verified the use of this Percentage-based Pattern (PHP) training approach using various comprehensive programs and applied this approach to genetic algorithm based curve fitting problems. The results in both cases were encouraging. The PHP-based hybrid algorithms resulted in significant reduction in the testing error as well as in the training time. The PHP algorithm is therefore concluded to be an approach towards more controlled learning algorithm in a field dominated by blind search methods.

Original languageEnglish
Pages (from-to)241-273
Number of pages33
JournalJournal of Intelligent Systems
Issue number3
Publication statusPublished - 2007
Externally publishedYes


  • Genetic algorithm
  • Hybrid training
  • Pattern learning
  • Supervised learning
  • Training parameters
  • Training pattern


Dive into the research topics of 'A lateral symmetry approach to Percentage-based Hybrid Pattern (PHP) training'. Together they form a unique fingerprint.

Cite this