A Comparative Analysis of Reinforcement Learning Methods
This paper analyzes the suitability of reinforcement learning (RL) for both programming and adapting situated agents. We discuss two RL algorithms: Q-learning and the Bucket Brigade. We introduce a special case of the Bucket Brigade, and analyze and compare its performance to Q in a number of...
| Main Author: | |
|---|---|
| Language: | en_US |
| Published: |
2004
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/1721.1/5978 |
| _version_ | 1826188877363150848 |
|---|---|
| author | Mataric, Maja |
| author_facet | Mataric, Maja |
| author_sort | Mataric, Maja |
| collection | MIT |
| description | This paper analyzes the suitability of reinforcement learning (RL) for both programming and adapting situated agents. We discuss two RL algorithms: Q-learning and the Bucket Brigade. We introduce a special case of the Bucket Brigade, and analyze and compare its performance to Q in a number of experiments. Next we discuss the key problems of RL: time and space complexity, input generalization, sensitivity to parameter values, and selection of the reinforcement function. We address the tradeoffs between the built-in and learned knowledge and the number of training examples required by a learning algorithm. Finally, we suggest directions for future research. |
| first_indexed | 2024-09-23T08:06:04Z |
| id | mit-1721.1/5978 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T08:06:04Z |
| publishDate | 2004 |
| record_format | dspace |
| spelling | mit-1721.1/59782019-04-09T16:39:11Z A Comparative Analysis of Reinforcement Learning Methods Mataric, Maja reinforcement learning situated agents inputsgeneralization complexity built-in knowledge This paper analyzes the suitability of reinforcement learning (RL) for both programming and adapting situated agents. We discuss two RL algorithms: Q-learning and the Bucket Brigade. We introduce a special case of the Bucket Brigade, and analyze and compare its performance to Q in a number of experiments. Next we discuss the key problems of RL: time and space complexity, input generalization, sensitivity to parameter values, and selection of the reinforcement function. We address the tradeoffs between the built-in and learned knowledge and the number of training examples required by a learning algorithm. Finally, we suggest directions for future research. 2004-10-04T14:25:16Z 2004-10-04T14:25:16Z 1991-10-01 AIM-1322 http://hdl.handle.net/1721.1/5978 en_US AIM-1322 13 p. 1444645 bytes 1130480 bytes application/postscript application/pdf application/postscript application/pdf |
| spellingShingle | reinforcement learning situated agents inputsgeneralization complexity built-in knowledge Mataric, Maja A Comparative Analysis of Reinforcement Learning Methods |
| title | A Comparative Analysis of Reinforcement Learning Methods |
| title_full | A Comparative Analysis of Reinforcement Learning Methods |
| title_fullStr | A Comparative Analysis of Reinforcement Learning Methods |
| title_full_unstemmed | A Comparative Analysis of Reinforcement Learning Methods |
| title_short | A Comparative Analysis of Reinforcement Learning Methods |
| title_sort | comparative analysis of reinforcement learning methods |
| topic | reinforcement learning situated agents inputsgeneralization complexity built-in knowledge |
| url | http://hdl.handle.net/1721.1/5978 |
| work_keys_str_mv | AT mataricmaja acomparativeanalysisofreinforcementlearningmethods AT mataricmaja comparativeanalysisofreinforcementlearningmethods |