Path-Local Learning in Reward-Modulated Tangled Program Graphs

Created by W.Langdon from gp-bibliography.bib Revision:1.8834

@InCollection{Naqvi:2026:raLGP,

• author = "Ali Naqvi and Stephen Kelly",
• title = "Path-Local Learning in Reward-Modulated Tangled Program Graphs",
• booktitle = "Recent Advances in Linear Genetic Programming",
• publisher = "Springer",
• year = "2026",
• editor = "Wolfgang Banzhaf and Ting Hu",
• chapter = "9",
• pages = "197--217",
• note = "forthcoming",
• keywords = "genetic algorithms, genetic programming, Linear Genetic Programming, TPG, Teams, Maze",
• abstract = "... especially in dynamic environments. Evolutionary computation typically selects policies by cumulative return, and Tangled Program Graphs follow this convention, with limited sensitivity to the temporal structure of reward. To better align the search process with step-wise feedback, we demonstrate two ways to expose immediate reward during decision-making: (1) appending reward to the observable state and (2) reward- modulated eligibility traces, implemented in a Tangled Program Graph variant with online learning-rule updates. the eligibility-trace approach substantially improves performance and produces policies in which fewer instructions are executed per timestep. This simple addition opens a pathway to integrate step-wise reward ...",
• notes = "part of \cite{Banzhaf:2026:raLGP_book}",

}

Genetic Programming entries for Ali Naqvi Stephen Kelly

Citations