2019 12.24

Building Dialog Systems with Less Supervision 讲座笔记

Lack of data is the number one challenge in deploying end-to-end trainable dialog systems for real-world applications. This talk will cover how to use learning methods to train a good model with less supervision. We will talk about how to integrate data augmentation, intermedia scaffolds, meta-learning to move towards the next-generation data-efficient dialog systems. We will briefly describe how to ensure the safety of the deployed system as well.... 387 0