Loading...

Cancer Simulation with Markov Decision Process

Zarepour, Fariborz | 2016

679 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 49407 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Habibi, Jafar
  7. Abstract:
  8. Cancer is refer to a class of diseases that create as the result of abnormal growth of cells and invasion of them to normal cells of human body, and annually cause the considerable percentage of death in the world. Because cancer can be considered as a complex system, various models presented to modeling and simulation of the behavior of it, using of different methods such as cellular automata, agent-based, game theory and other methods. Multi-agent simulation models as a special kind of agent-based models, is a method that used to simulate some real-world phenomena that usually contains many different components and interact using different and complex ways. Since the cells are located in an interactive environment and cell actions can be viewed as a series of sequential actions, markov decision process can be a proper approach in order to create the behavioral model of cells in cancer tumors, for select the their behaviors in the environment. In absence of a complete Markov decision process model of the environment, Q-learning as one of the most important methods of reinforcement learning can be used to find the optimal action selection policy and by simulating the problem environment, can create the behavioral models of environment agents. In the proposed model, Q-learning method creates behavioral models of the environment agents and using these models, tumor growth is simulated. Validation the results of the proposed model with mathematical models and experimental observation of tumor growth, shows that the Q-learning method can be a good way to simulate the behavior of the agents in the tumor environment
  9. Keywords:
  10. Reinforcement Learning ; Q-Learning ; Cancer Simulation ; Markov Decision Making ; Multi-Agent Markov Chain Processes (MMDPs) ; Multi-Agent Reinforcement Learning

 Digital Object List

 Bookmark

No TOC