“School of Cognitive”

Back to Papers Home
Back to Papers of School of Cognitive

Paper   IPM / Cognitive / 11397
School of Cognitive Sciences
  Title:   Concept Learning using Reinforcement Learning Based on Evidence theory
  Author(s): 
1.  Taher Shahbazi
2.  Majid Nili Ahmadabadi
3.  Babak Nadjar Arabi
  Status:   In Proceedings
  Proceeding: 1st Joint Congress on Fuzzy and Intelligent Systems, Mashhad, Iran, 2007
  Year:  2007
  Supported by:  IPM
  Abstract:
Due to the complicated nature of concept learning problem and limitations in the real environments, it is inevitable for the learning agent to model and deal with uncertainty. In this paper, we take evidence theory into consideration for modeling uncertainty in concept learning. Belief function framework, as a mapping from agent?s continuous sensory space to the discrete actions space, constructs an evidential model for reasoning under uncertainty during learning process. The agent learns this evidential mapping through interaction with the environment or mentor, using reinforcement signal. We show the way of learning these belief functions using reinforcement learning and evaluate the method by a synthetic simulator.

Download TeX format
back to top
scroll left or right