Name
Abstract | ||
|---|---|---|
The field of general intelligence is one, where humans can still easily outperform machines. In the context of our work we describe it as the ability to learn an activity, like playing a game, without any prior knowledge of goals and rules. The agent has to learn by doing/playing and examining the consequences of its actions. Many traditional techniques in reinforcement learning, such as SARSA and Q-Learning, can provide a good solution to this category of problems. In our paper, however, we propose an alternative method based on evolutionary algorithms to overcome the extensive computing for all state-action pairs needed in traditional approaches. We have evaluated various parent selection algorithms and two different fitness functions. "The General Video Game AI Competition" (GVGAI), where contestants submit a playing agent programmed with some learning algorithm to be tested against unknown games, has been used as a benchmark for the performance of our implementation. |
Year | DOI | Venue |
|---|---|---|
2019 | 10.1007/978-3-030-14799-0_27 | INTELLIGENT INFORMATION AND DATABASE SYSTEMS, ACIIDS 2019, PT I |
Keywords | Field | DocType |
Artificial intelligence, General intelligence, Planning agents, Video game controllers | Control theory,Evolutionary algorithm,Computer science,Artificial intelligence,Machine learning,Reinforcement learning | Conference |
Volume | ISSN | Citations |
11431 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 0 | 2 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Kristiyan Balabanov | 1 | 1 | 1.04 |
| Doina Logofatu | 2 | 17 | 16.74 |