A generalised dropout mechanism for distributed systems

Bull, Larry; Liu, Haixia

Abstract

This letter uses a modified form of the NK model introduced to explore aspects of distributed control. In particular, a previous result suggesting the use of dynamically formed subgroups within the overall system can be more effective than global control is further explored. The conditions under which the beneficial distributed control emerges are more clearly identified, and the reason for the benefit over traditional global control is suggested as a generally applicable dropout mechanism to improve learning in such systems.

• Journal Article Type: Article
• Acceptance Date: May 13, 2022
• Online Publication Date: Oct 24, 2022
• Publication Date: May 1, 2023
• Deposit Date: May 20, 2022
• Publicly Available Date: Nov 25, 2022
• Journal: Artificial life
• Print ISSN: 1064-5462
• Electronic ISSN: 1530-9185
• Publisher: Massachusetts Institute of Technology Press (MIT Press)
• Peer Reviewed: Peer Reviewed
• Volume: 29
• Issue: 2
• Pages: 146-152
• DOI: https://doi.org/10.1162/artl_a_00393
• Keywords: Rugged fitness landscape; multi-agent systems; neural networks; NKD model; search
• Public URL: https://uwe-repository.worktribe.com/output/9541079
• Publisher URL: https://direct.mit.edu/artl/article-abstract/doi/10.1162/artl_a_00393/113523/A-Generalised-Dropout-Mechanism-for-Distributed?redirectedFrom=fulltext