arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2102.13089cs.LG
收藏

辅助任务对表征动力学的影响

On The Effect of Auxiliary Tasks on Representation Dynamics

University of Oxford(牛津大学);DeepMind(深度思维公司)

Clare Lyle, Mark Rowland, Georg Ostrovski, Will Dabney

摘要:

虽然辅助任务在塑造强化学习代理所学习的表示方面起着关键作用,但关于实现这一点的机制仍有很多未知之处。这项工作发展了我们的理解之间的关系,辅助任务,环境结构,并表示通过分析动态的时间差算法。通过这种方法,我们建立了过渡算子的谱分解和各种辅助任务引起的表示之间的联系。然后,我们利用这些理论结果的见解,为稀疏奖励环境中的深度强化学习代理选择辅助任务提供信息。

英文摘要:

While auxiliary tasks play a key role in shaping the representations learnt by reinforcement learning agents, much is still unknown about the mechanisms through which this is achieved. This work develops our understanding of the relationship between auxiliary tasks, environment structure, and representations by analysing the dynamics of temporal difference algorithms. Through this approach, we establish a connection between the spectral decomposition of the transition operator and the representations induced by a variety of auxiliary tasks. We then leverage insights from these theoretical results to inform the selection of auxiliary tasks for deep reinforcement learning agents in sparse-reward environments.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑