Program Synthesis on Single-Layer Loop Behavior in Pure Functional Programming
Created by W.Langdon from
gp-bibliography.bib Revision:1.7975
- @InProceedings{hsu:2024:CEC,
-
author = "Tzu-Hao Hsu and Chi-Hsien Chang and Tian-Li Yu",
-
title = "Program Synthesis on Single-Layer Loop Behavior in
Pure Functional Programming",
-
booktitle = "2024 IEEE Congress on Evolutionary Computation (CEC)",
-
year = "2024",
-
editor = "Bing Xue",
-
address = "Yokohama, Japan",
-
month = "30 " # jun # " - 5 " # jul,
-
publisher = "IEEE",
-
keywords = "genetic algorithms, genetic programming, Synthesizers,
Redundancy, Search problems, Feature extraction, Market
research, Maintenance, Program Synthesis",
-
isbn13 = "979-8-3503-0837-2",
-
DOI = "doi:10.1109/CEC60901.2024.10612128",
-
abstract = "Program synthesis (PS) is a field devoted to
auto-matically generating computer programs from
high-level specifications, and genetic programming (GP)
is one commonly-used way to achieve PS. PushGP,
operating on a stack-based language, is considered as a
state-of-the-art program synthesiser among GPs, while
another research trend foucus on the grammar-based
languages due to the readability and the ease of
maintenance. In this paper, we propose the repetitive
structure genetic programming (RSGP), a new
grammar-based program synthesiser under the pure
functional programming paradigm. RSGP defines a
recursive function to simulate the single-layer loop
behaviour and leverages the minimum redundancy maximum
relevance $(\text{mRMR})$ feature selection with the
Pearson correlation coefficient (PCC) to select the
capable and diverse programs for the next generation.
The experiment results show that RSGP outperforms
PushGP, CBGP, and HOTGP in terms of the number of
successful programs on CountOdds and LastIndexofZero
from PSBl, Luhn from PSB2, and 3 out of 4 designed
problems. Additionally, the ablation study indicates
that using $\text{mRMR}$ with PCC does encourage proper
problem decomposition with the trade-off of diminishing
the search ability within a similar neighbourhood. RSGP
uses an adaptation mechanism to balance the trade-off
to automatically fit the needs of different problems.",
-
notes = "also known as \cite{10612128}
WCCI 2024",
- }
Genetic Programming entries for
Tzu-Hao Hsu
Chi-Hsien Chang
Tian-Li Yu
Citations