Sample-efficient Actor-Critic Reinforcement Learning with Supervised Data for Dialogue Management
Pei-Hao Su, Paweł Budzianowski, Stefan Ultes, Milica Gašić, Steve Young
Abstract
Deep reinforcement learning (RL) methods have significant potential for dialogue policy optimisation. However, they suffer from a poor performance in the early stages of learning. This is especially problematic for on-line learning with real users. Two approaches are introduced to tackle this problem. Firstly, to speed up the learning process, two sample-efficient neural networks algorithms: trust region actor-critic with experience replay (TRACER) and episodic natural actor-critic with experience replay (eNACER) are presented. For TRACER, the trust region helps to control the learning step size and avoid catastrophic model changes. For eNACER, the natural gradient identifies the steepest ascent direction in policy space to speed up the convergence. Both models employ off-policy learning with experience replay to improve sample-efficiency. Secondly, to mitigate the cold start issue, a corpus of demonstration data is utilised to pre-train the models prior to on-line reinforcement learning. Combining these two approaches, we demonstrate a practical approach to learn deep RL-based dialogue policies and demonstrate their effectiveness in a task-oriented information seeking domain.- Anthology ID:
- W17-5518
- Volume:
- Proceedings of the 18th Annual SIGdial Meeting on Discourse and Dialogue
- Month:
- August
- Year:
- 2017
- Address:
- Saarbrücken, Germany
- Editors:
- Kristiina Jokinen, Manfred Stede, David DeVault, Annie Louis
- Venue:
- SIGDIAL
- SIG:
- SIGDIAL
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 147–157
- Language:
- URL:
- https://aclanthology.org/W17-5518
- DOI:
- 10.18653/v1/W17-5518
- Cite (ACL):
- Pei-Hao Su, Paweł Budzianowski, Stefan Ultes, Milica Gašić, and Steve Young. 2017. Sample-efficient Actor-Critic Reinforcement Learning with Supervised Data for Dialogue Management. In Proceedings of the 18th Annual SIGdial Meeting on Discourse and Dialogue, pages 147–157, Saarbrücken, Germany. Association for Computational Linguistics.
- Cite (Informal):
- Sample-efficient Actor-Critic Reinforcement Learning with Supervised Data for Dialogue Management (Su et al., SIGDIAL 2017)
- PDF:
- https://preview.aclanthology.org/bionlp-24-ingestion/W17-5518.pdf