Evolving parsimonious circuits through shapley value-based genetic programming

Xinming SHI, Jiashi GAO, Leandro L. MINKU, Xin YAO

Research output: Book Chapters | Papers in Conference ProceedingsConference paper (refereed)Researchpeer-review

Abstract

Evolutionary analog circuit design is a challenging task due to the large search space incurred by the circuit topology and device values. Applying genetic operators on randomly selected genes may make it difficult to identify which part of sub-circuit is beneficial to the evolution and even destroy useful sub-circuits, potentially incurring stagnation of the evolutionary process and bloat on the evolved circuits. In this paper, we propose a tree-based approach called Shapley Circuit Tree that incorporates Shapley values for quantifying the contribution of each function node of the circuit tree to the performance of the whole tree, to guide the evolutionary process. Our experiments on three benchmarks show that the proposed approach is able to evolve analog circuits with smaller area while converging faster than existing approaches. © 2022 Owner/Author.
Original languageEnglish
Title of host publicationGECCO 2022 Companion - Proceedings of the 2022 Genetic and Evolutionary Computation Conference
PublisherAssociation for Computing Machinery, Inc
Pages602-605
Number of pages4
ISBN (Print)9781450392686
DOIs
Publication statusPublished - 9 Jul 2022
Externally publishedYes

Bibliographical note

This work was support by the Research Institute of Trustworthy Autonomous Systems (RITAS), the Guangdong Provincial Key Laboratory (Grant No. 2020B121201001), the Program for Guangdong Introducing Innovative and Enterpreneurial Teams (Grant No. 2017ZT07X386), the Shenzhen Science and Technology Program (Grant No. KQTD2016112514355531).

Keywords

  • evolutionary analog circuit design
  • evolvable hardware
  • genetic programming
  • shapley value
  • tree-based circuit representation

Fingerprint

Dive into the research topics of 'Evolving parsimonious circuits through shapley value-based genetic programming'. Together they form a unique fingerprint.

Cite this