Please use this identifier to cite or link to this item: https://hdl.handle.net/10356/47027
Title: Q-learning algorithms using fuzzy inference system
Authors: Kattur Rajarethinam Ganeshrajhan
Keywords: DRNTU::Engineering
Issue Date: 2009
Abstract: Reinforcement Learning is the learning algorithm that learns by direct experimentation. It works by maximizing the numeric reward received from the environment. Out of the famous Reinforcement Algorithms, the most popular one is the Q-Learning algorithm. It stores a Qvalue with respect to the quality of the action performed for a specific state. It has a limitation of generalization in continuous environment. This limitation is solved by adopting the Fuzzy Inference System in Fuzzy Q-Learning (FQL). Since FQL only involves parameter modification only, structure identification is done in Dynamic Fuzzy Q-Learning (DFQL) by automatic generation of fuzzy agents. Although the DFQL is an online algorithm where rules are generated automatically, there is a possibility that some rules can become redundant. Hence, pruning of redundant rules, which is adopted in Dynamic Self-Generated Fuzzy QLearning (DSGFQL) is proposed. An Extended Self Organizing Map was added to the DSGFQL algorithm and a new algorithm Enhanced Dynamic Self-Generated Fuzzy QLearning (EDSGFQL) scheme is proposed. Another method replacing the structure identification method in DFQL by Incremental proposed Topological Preserving Map approach was proposed. The scheme is known as Incremental-Topological-Preserving-Map- Based Fuzzy Q-Learning. These methods involve Q-Learning where continuous actions were not generated. Another method that can generate continuous action known as Continuous Action Q-Learning was proposed. Implementing the FIS in the continuous action Q-Learning will give Fuzzy Continuous Action Q-Learning.
Description: 53 p.
URI: http://hdl.handle.net/10356/47027
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
EEE_THESES_78.pdf
  Restricted Access
5.92 MBAdobe PDFView/Open

Google ScholarTM

Check

Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.