Conference Papers Year : 2022

Pre-loaded Deep-Q Learning

Abstract

This paper explores the potentiality of pre-loading deep-Q learning agents’ replay memory buffers with experiences generated by preceding agents, so as to bolster their initial performance. The research illustrates that this pre-loading of previously generated experience replays does indeed improve the initial performance of new agents, provided that an appropriate degree of ostensibly undesirable activity was expressed in the preceding agent’s behaviour.
Fichier principal
Vignette du fichier
529309_1_En_14_Chapter.pdf (835.72 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04178733 , version 1 (08-08-2023)

Licence

Identifiers

Cite

Tristan Falck, Elizabeth Ehlers. Pre-loaded Deep-Q Learning. 12th International Conference on Intelligent Information Processing (IIP), May 2022, Qingdao, China. pp.159-172, ⟨10.1007/978-3-031-03948-5_14⟩. ⟨hal-04178733⟩
31 View
5 Download

Altmetric

Share

More