|
The Study of Emotion, Learning and Intelligence |
|
Report Number: |
UMASSD-CIS-TR-2006010 |
|
Publication Type: |
Unpublished |
|
File Name: |
UMASSD-CIS-TR-2006010.pdf
|
|
Abstract: |
Emotion has been shown to be an important factor in a human???s decision making and learning processes. Affective computing is to simulate the human???s emotion mechanism with computer programs. Meanwhile, reinforcement learning is an advanced machine learning technique that has attracted a lot of attention recently. However, it has difficulty learning in a dynamic world or in a large state space. The general goal of our study is to discover the relationships between emotion and learning. The first step is to find whether and how affective computing mechanism can be applied to reinforcement learning and help it overcome some if its limitations. We created two agent architectures, one based on affective computing and the other on active reinforcement learning, where we use some abstraction and modeling techniques driven from the affective computing to build the world model for the learning agent. Experiments have been performed to study the performance of both the affective computing agent and the modified reinforcement learning agent in a large, dynamic grid world environment. |
|
Authors: |
Jason Williams
(Primary Contact)
The Study of Emotion, Learning and Intelligence
UMAss Dartmouth
CIS
g_jwilliams@umassd.edu
|
|
