arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1808.09442cs.CLcs.AIcs.LG
收藏

判别式Deep Dyna-Q:对话策略学习的稳健规划

Discriminative Deep Dyna-Q: Robust Planning for Dialogue Policy Learning

Microsoft Research(微软研究院);National Taiwan University(台湾大学)

Shang-Yu Su, Xiujun Li, Jianfeng Gao, Jingjing Liu, Yun-Nung Chen

摘要:

本文提出了一种判别式Deep Dyna-Q(D3 Q)方法,以提高Deep Dyna-Q(DDQ)的有效性和鲁棒性,DDQ是最近提出的一种框架,扩展了Dyna-Q算法,以集成任务完成对话策略学习的规划。为了克服DDQ对模拟体验质量的高度依赖性,我们在D3 Q中引入了一个基于RNN的训练器,以区分模拟体验和真实用户体验,从而控制训练数据的质量。实验表明,D3 Q通过控制用于规划的模拟经验的质量,显著优于DDQ。D3 Q的有效性和鲁棒性进一步证明了在域扩展设置,其中代理的能力,以适应不断变化的环境进行了测试。

英文摘要:

This paper presents a Discriminative Deep Dyna-Q (D3Q) approach to improving the effectiveness and robustness of Deep Dyna-Q (DDQ), a recently proposed framework that extends the Dyna-Q algorithm to integrate planning for task-completion dialogue policy learning. To obviate DDQ's high dependency on the quality of simulated experiences, we incorporate an RNN-based discriminator in D3Q to differentiate simulated experience from real user experience in order to control the quality of training data. Experiments show that D3Q significantly outperforms DDQ by controlling the quality of simulated experience used for planning. The effectiveness and robustness of D3Q is further demonstrated in a domain extension setting, where the agent's capability of adapting to a changing environment is tested.

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。

↑