PHD Research Proposal in Reinforcement Learning

Reinforcement learning is one of the learning paradigms in the machine learning, whereas a learning agent interacts with the environment and, perceiving the consequences of their actions, can learn to change over its behavior concerning rewards acquired [1]. By leveraging the procured rewards, the learning agent seeks to enhance their knowledge about the environment for determining future action. The principal target of reinforcement learning is to decide the best sequence of action for each given state in the environment for maximizing the cumulative reward. With the abundant, successful applications in gaming, plant control, and business intelligence, the reinforcement learning technique is considered as ideal for decision making in obscure models or with the unknown environment
It is also suitable for the may real-time application such as personalized web services, finance sector, PC games, inventory management, robotics in the industrial automation, traffic light control, bidding, and advertising, and so on [2].
Reinforcement learning is precisely useful in various applications, despite it faces the ridiculous challenging issues [3]. Recently, in the reinforcement learning that accomplishing multi-task learning is the major challenge, whereas the critical point of the issues is the scalability. Another challenge in reinforcement learning is a safe and effective exploration.
Moreover, the trade-off between the exploration and exploitation and long term credit assignment remains a stumbling block. The computational power required by the reinforcement learning hinders several techniques including exhaustive search, tabular methods, and so on.
Even though, the reinforcement learning confronts the challenges in stability, convergence and optimality analysis. Moreover, it entails the additional mechanism to improve the learning efficiency of the system Notably, it considered as inappropriate for taking the best actions for the real-time environment, whereas the dynamics in the environment varies too many times and becomes a great deal of money for the robot to learn for circumventing the aforementioned problem.

Reference:

  • [1] Sutton, Richard S., and Andrew G. Barto, “Reinforcement learning: An introduction”, 2011.

  • [2] Hou, Jun, Hua Li, Jinwen Hu, Chunhui Zhao, Yaning Guo, Sijia Li, and Quan Pan, “A review of the applications and hotspots of reinforcement learning”, In IEEE International Conference on Unmanned Systems (ICUS), pp.506-511, 2017.

  • [3] Kormushev, Petar, Sylvain Calinon, and Darwin Caldwell, “Reinforcement learning in robotics: Applications and real-world challenges” Robotics, Vol2, No.3, pp.122-148, 2013.

Leave Comment

Your email address will not be published. Required fields are marked *

clear formSubmit