Dissertation > Excellent graduate degree dissertation topics show

Model-based dynamic hierarchical reinforcement learning algorithm

Author: YuanZuoHong
Tutor: WuMin；ChenZuo
School: Central South University
Course: Control Science and Engineering
Keywords: Agent Model-based reinforcement learning Adaptive Clustering Dynamic Hierarchical Reinforcement Learning MAXQ
CLC: TP181
Type: Master's thesis
Year: 2011
Downloads: 56
Quote: 0
Read: Download Dissertation

Abstract


Reinforcement learning due to self-learning and online learning the good characteristics of the field of machine learning has become an important branch. However, the agent in making large-scale high-dimensional environment for enhanced learning by \unable to complete the learning task. Therefore, if we can effectively alleviate the \to promote the field of machine learning theory and technology development of great significance. Therefore, in order to alleviate the unknown scale environment \explore information based adaptive clustering algorithm for dynamic hierarchical reinforcement learning. The algorithm dynamically generates fusion tense state abstraction and abstract (or motion abstract) The MAXQ hierarchical structure, thus limiting the MAXQ by each subtask Strategy Search space significantly faster learning speed. First, the model-based reinforcement learning process, using information based on adaptive clustering algorithm to explore the entire state space is divided into several states subspaces, i.e. the status of the task abstraction done automatically stratification and the subspace based on state Termination of state set, made - kind of an improved action selection strategy. Secondly, according to the frequency of the effective implementation of the action tense situation similar to MAXQ automatically generated abstract hierarchical structure, and thus the effective set of actions will be based on each state subspace MAXQ classified to the appropriate sub-tasks, thus automatically generate fusion of state abstraction and temporal abstraction MAXQ hierarchy. Again, based on the hierarchical framework MAXQ task recursive optimal search strategies and learning process in the future MAXQ architecture dynamically adjusted to reduce the initial hierarchy unreasonable limitations. Through simulation experiments show that the proposed algorithm can significantly improve the unknown environment-agent learning efficiency, effectively alleviate the \Finally, the papers summarize and make some further study the issue.

Related Dissertations

  1. The Problems and Countermeasures in Respect to Agent Construction of the Project: Guangzhou Asian Games Venues,G812.2
  2. "Social Agent": between Active and Inactive,D625
  3. Research and Implementation on Service-Oriented Multi-Agent System Cooperation Mechanism,TP393.09
  4. The Preparation of Aluminum-Ammonium Cationic Starch and Its Application in Papermaking,TS727
  5. Research of Communication Mechanism in the Distributed Network Based on Mobile Agent,TP393.02
  6. The optimal scale of investment of private equity funds,O224
  7. Preliminary Study on Obstacles Effect of Succession Cropping and Reduction Techniques of Microbial Agents in Flue-cured Tobacco,S572
  8. Effect of Flue-cured Tobacco Continuous Cropping on Soil Chemical and Physical Properties and Tobacco Leaf Quality,S572
  9. Development and Application of Biological Seed Coating Formulation for Soybean,S565.1
  10. The Application of Different Particle Size Peanut Shell Added by Wetting Agent and Nutrient Solution in Cucumber Nursery,S642.2
  11. The Research on Localization Algorithm of Wireless Sensor Networks,TN929.5
  12. Research on Agent Construction System Management Model of Highway Project,F283
  13. A Research on Soil Transmission Medium of Rice Stripe Virus,S435.111.42
  14. A Research on Soil Transmission Medium of Rice Stripe Virus,S435.111.42
  15. Study of the Water-Inrush Mechanism and Prediction of the Water-Resisting Floor,TQ133.1
  16. Experimental Study of Wettability of Coal and New Wetting Agent,TD714
  17. Research on the Plasticizers of Starch Sizing Agents,TS103.846.1
  18. Survey and Research Analysis of Leavening Agents in Food Application,TS202.3
  19. Development of ATP Bioluminescent Method and Its Kit for the Detection of Bacterial Count in Food,TS207.4
  20. Analysis of Efficacy of Radiotherapy and 5-FU Slow Release Particles in Treating Non-small Cell Lung Cancer,R734.2
  21. Heping District Library project agent system implementation phase of the project management optimization studies,G251

CLC: > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Automated reasoning,machine learning
© 2012 www.DissertationTopic.Net  Mobile