arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2003.13661cs.LGcs.AIcs.ROstat.ML
收藏

软模块化多任务强化学习

Multi-Task Reinforcement Learning with Soft Modularization

UC San Diego(加州大学圣迭戈分校);UC Berkeley(加州大学伯克利分校);IIIS, Tsinghua(清华大学交叉信息研究院)

Ruihan Yang, Huazhe Xu, Yi Wu, Xiaolong Wang

摘要:

多任务学习是强化学习中一个非常具有挑战性的问题。虽然联合训练多个任务允许策略在不同任务之间共享参数,但优化问题变得不简单:仍然不清楚网络中的哪些参数应该在任务之间重用,以及来自不同任务的梯度如何相互干扰。因此,而不是天真地在任务之间共享参数,我们引入了一个明确的模块化技术的政策表示,以减轻这个优化问题。给定一个基本策略网络,我们设计了一个路由网络,估计不同的路由策略,以重新配置每个任务的基本网络。我们的任务特定策略不是直接为每个任务选择路由,而是使用一种称为软模块化的方法来软组合所有可能的路由,这使得它适用于顺序任务。我们在仿真中对各种机器人操作任务进行了实验,并表明我们的方法在强基线上大幅提高了样本效率和性能。

英文摘要:

Multi-task learning is a very challenging problem in reinforcement learning. While training multiple tasks jointly allow the policies to share parameters across different tasks, the optimization problem becomes non-trivial: It remains unclear what parameters in the network should be reused across tasks, and how the gradients from different tasks may interfere with each other. Thus, instead of naively sharing parameters across tasks, we introduce an explicit modularization technique on policy representation to alleviate this optimization issue. Given a base policy network, we design a routing network which estimates different routing strategies to reconfigure the base network for each task. Instead of directly selecting routes for each task, our task-specific policy uses a method called soft modularization to softly combine all the possible routes, which makes it suitable for sequential tasks. We experiment with various robotics manipulation tasks in simulation and show our method improves both sample efficiency and performance over strong baselines by a large margin.

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。

↑