Please use this identifier to cite or link to this item:
https://olympias.lib.uoi.gr/jspui/handle/123456789/10956Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.author | Likas, A | en |
| dc.contributor.author | Kontoravdis, D. | en |
| dc.contributor.author | Stafylopatis, A. | en |
| dc.date.accessioned | 2015-11-24T17:01:38Z | - |
| dc.date.available | 2015-11-24T17:01:38Z | - |
| dc.identifier.uri | https://olympias.lib.uoi.gr/jspui/handle/123456789/10956 | - |
| dc.rights | Default Licence | - |
| dc.title | Enhancing stochasticity in reinforcement learning schemes | en |
| heal.type | journalArticle | - |
| heal.type.en | Journal article | en |
| heal.type.el | Άρθρο Περιοδικού | el |
| heal.language | en | - |
| heal.access | campus | - |
| heal.recordProvider | Πανεπιστήμιο Ιωαννίνων. Σχολή Θετικών Επιστημών. Τμήμα Μηχανικών Ηλεκτρονικών Υπολογιστών και Πληροφορικής | el |
| heal.publicationDate | 1995 | - |
| heal.abstract | The paper develops reinforcement algorithms for networks of stochastic units which select their output based on a distribution whose dependence on the controllable parameters (weights) of the network is not deterministic. A special case of the proposed schemes concerns those applied to Normal/Bernoulli units, which are binary units with two stochastic levels. Both REINFORCE algorithms as well as algorithms not belonging to the REINFORCE class have been developed. All schemes are designed to exploit the two parameters of a normal distribution in order to explore discrete domains. The ability of the proposed algorithms to perform efficient exploration is tested in a number of optimization problems concerning the maximization of a set of functions defined on binary domains. Particular emphasis has been given on deriving schemes having the property of sustained exploration. Obtained results indicate the superiority of the reinforcement schemes applied to Normal/Bernoulli units over reinforcement schemes applied to single-parameter Bernoulli units. | en |
| heal.journalName | Journal of Intelligent Systems | en |
| heal.journalType | peer reviewed | - |
| heal.fullTextAvailability | TRUE | - |
| Appears in Collections: | Άρθρα σε επιστημονικά περιοδικά ( Ανοικτά) | |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| Likas-1995-Enhancing stochasticity in reinforcement learning schemes.pdf | 1.86 MB | Adobe PDF | View/Open Request a copy |
This item is licensed under a Creative Commons License
