持续的深度强化学习与任务无意识的政策蒸
Muhammad Burhan Hafez1, Kerim Erekmen2
1School of Electronics and Computer Science, University of Southampton, Southampton, SO17 1BJ, United Kingdom. burhan.hafez@soton.ac.uk.
本研究介绍了任务无意识的政策蒸 (TAPD),这是持续学习系统的新框架. TAPD使代理人能够有效地学习新任务而不会忘记旧任务,从而提高整体性能和可扩展性.
科学领域:
- 人工智能的人工智能
- 机器学习 机器学习
- 机器人技术 机器人技术 机器人技术
背景情况:
- 持续学习系统的目的是解决多重任务,而无需再培训.
- 灾难性遗忘,缺乏积极转移,可扩展性和未标记的数据是关键的挑战.
- 目前的方法需要大量的培训时间来完成每一项新任务.
研究的目的:
- 引入一个新的框架,任务无意识的政策蒸 (TAPD),以解决持续学习的局限性.
- 为了使代理人能够有效地学习新的任务,而不会忘记以前获得的知识.
- 提高在通用学习系统中的样本效率和可扩展性.
主要方法:
- 拟议的任务无关政策蒸 (TAPD) 框架包含了一个任务无关的探索阶段.
- 代理人在探索过程中最大限度地提高了内在的动机,以自我监督的方式寻求新的状态.
- 在任务不可知阶段获得的知识被提炼为有效的下游任务学习.
主要成果:
- 塔普德框架有效地缓解了灾难性遗忘,并增强了积极的前期转移.
- 这种方法在众多任务中显示出更好的可扩展性.
- 经过TAPD培训的特工在解决下游任务时显著提高了样本效率.
结论:
- 任务无意识的政策蒸 (TAPD) 为持续学习挑战提供了有效的解决方案.
- 自主监督的,任务无关的探索能够有效地转移知识和适应.
- 塔普德推进了更强大,更可扩展的通用学习系统的开发.
更多相关视频
08:05A Prediction Error-driven Retrieval Procedure for Destabilizing and Rewriting Maladaptive Reward Memories in Hazardous Drinkers
Published on: January 5, 2018
11:09RBDT: A Computerized Task System based in Transposition for the Continuous Analysis of Relational Behavior Dynamics in Humans
Published on: July 17, 2021
相关概念视频
Generalization, Discrimination, and Extinction
Generalization occurs when a behavior reinforced in one context is performed in similar situations. For instance, a student who studies diligently for calculus and receives excellent grades might apply the same study habits to psychology and history, expecting similar results. Generalization shows how learning in one setting can influence behavior in...
Reinforcement Schedules
Once a behavior is learned,...
Associative Learning
Classical conditioning, also known...
Role of Shaping in Operant Conditioning
The steps involved in shaping begin with reinforcing any response that resembles the desired behavior. For example, parents might praise a child for picking up one toy. As...
Distribution Reliability and Automation
Long-term Potentiation
