NATIONAL BIBLIOGRAPHY NUMBER
Number
۱۹۸۱۱پ
LANGUAGE OF THE ITEM
.Language of Text, Soundtrack etc
per
TITLE AND STATEMENT OF RESPONSIBILITY
Title Proper
یادگیری تقویتی چندعاملی براساس به اشتراکگذاری وزندار تجارب بهصورت تقاضامحور
Parallel Title Proper
Multi-Agent Reinforcement Learning Based on On-Demand Weighted Experience Sharing
First Statement of Responsibility
/رویا قاسمزاده
.PUBLICATION, DISTRIBUTION, ETC
Name of Publisher, Distributor, etc.
: مهندسی برق و کامپیوتر
Date of Publication, Distribution, etc.
، ۱۳۹۷
Name of Manufacturer
، افشار
PHYSICAL DESCRIPTION
Specific Material Designation and Extent of Item
۶۸ص
NOTES PERTAINING TO PUBLICATION, DISTRIBUTION, ETC.
Text of Note
چاپی - الکترونیکی
DISSERTATION (THESIS) NOTE
Dissertation or thesis details and type of degree
ارشد
Discipline of degree
مهندسی کامپیوتر گرایش هوش مصنوعی
Date of degree
۱۳۹۷/۰۶/۲۰
Body granting the degree
تبریز
SUMMARY OR ABSTRACT
Text of Note
.Multi-agent systems are distributed systems of independent actors called agents. These systems are able to solve problems that single-agent systems are not capable of. Multi-agent reinforcement learning enables multi-agent systems to learn how to act in a complex environment without prior knowledge. One of the problems in multi-agent reinforcement learning systems is the possibility of improving learning process through agent's interactions. Sharing instantaneous information, sharing episodes and sharing learned policies are some methods of interactions among agents. In this thesis we propose a method for sharing learned policies in simultaneously learning agents that use reinforcement learning, more specifically Q-learning. Whenever an agent wants to choose an action, in case of not having enough experience, asks other agents to share their Q-values in that state. other agents in case of having more experience compared to requesting agent, send a list of Q-values of all possible actions in that state along with their confidence of that values to requesting agent. The requesting agent computes the weighted average of received values and updates its Q-table. Then chooses an action based on its action selection policy
PARALLEL TITLE PROPER
Parallel Title
Multi-Agent Reinforcement Learning Based on On-Demand Weighted Experience Sharing
TOPICAL NAME USED AS SUBJECT
spets rewef ni edosipe hcae evlos nac ,egareva no ,dohtem desoporp ,sedosipe gninrael gniruD .retsaf hcum etats taht sehcaer dohtem desoporp tub ecnamrofrep ralimis ot egrevnoc yllanif smhtirogla htoB .gninrael etarelecca dna ecneirepxe s'rehto hcae esu nac stnega edosipe gninael emos retfa tub ,gninrael tnemecrofnier tnega-elgnis ot ralimis skrow dohtem desoporp ruo sedosipe gninrael reilrae ni taht wohs stluser latnemirepxE
PERSONAL NAME - PRIMARY RESPONSIBILITY
قاسمزاده، رویا
Ghasemzadeh,Roya
نمایهسازی قبلی
