BROWSE
Detail
Conference Paper
Experience Augmentation for Fast Deep Reinforcement Learning
Cited - time in 
Share 
Share
- Authors
- Mingyu Pi, Moonyoung Chung
- Issue Date
- 2020-12
- Citation
- International Conference on Internet (ICONI) 2020, pp.1-2
- Language
- English
- Type
- Conference Paper
- Abstract
- The advantage of reinforcement learning is that it learns by itself without human intervention and background knowledge. The RL agent selects the optimal action based on rewards from their own experience and exploration, but because of this characteristic, takes a long time to train. While humans can learn similar experiences through one experience, machines can only learn about the experiences they have acted on. In this paper, we present a method to fast reinforcement learning using experience augmentation. We apply our method to the Cartpole environment and find that training time significantly reduces in both the DDQN(Double DQN) and 3DQN(Dueling Double DQN) algorithms.
- KSP Keywords
- Deep reinforcement learning, Long Time, Reinforcement Learning(RL), Training time, background knowledge, human intervention