Issue No. 04 - July/August (2003 vol. 18)
Paul Juell , North Dakota State University
Patrick Paulson , North Dakota State University
<p>The reinforcement-trained case-based reasoning system uses reinforcement learning to adjust its similarity metric. RETCBR learns similarity metrics in response to feedback from the user or environment, letting the system adapt to user needs. The system also works in application domains with insufficient expertise to develop similarity metrics, expanding the type of problems to which CBR can be applied. The authors describe two methods of implementing similarity assessment and demonstrate the methods' performance in a weather forecasting application. The weather application uses reports obtained from actual weather stations and demonstrates that the system can operate without explicit domain knowledge. The experiments show that CBR systems can improve their similarity metrics through reinforcement-learning techniques.</p>
neural networks, reasoning, learning, case-based reasoning, reinforcement learning
P. Paulson and P. Juell, "Using Reinforcement Learning for Similarity Assessment in Case-Based Systems," in IEEE Intelligent Systems, vol. 18, no. , pp. 60-67, 2003.