Continual Learning Dialogue Systems
- Learning on the Job after Model Deployment

Tutorial @ IJCAI 2021, August 21-26, 2021
Sahisnu Mazumder and Bing Liu



Second Edition: "Lifelong Machine Learning." by Z. Chen and B. Liu, Morgan & Claypool, August 2018 (1st edition, 2016).
  • Added three new chapters: (4) Continual Learning and Catastrophic Forgetting, (5) Open-world Learning, (8) Continuous Knowledge Learning in Chatbots
  • Introduced the concept of learning on the job or learning after deployment.
  • Updated and/or reorganized the other chapters.
  • Download the first edition, Lifelong Machine Learning, Nov 2016.
  • Two-Sentence Summary

    The tutorial focuses on the new research topic of building the next-generation dialogue systems (or chatbots) that can continuously and interactively learn from end-users during conversation (on-the-job after model deployment) to become more and more powerful over time. We will give the motivation and background of the topic and discuss existing techniques and open challenges, which we believe, will shape the future of dialogue systems development.

    Abstract

    Dialogue systems, also known as chatbots, have gained escalating popularity in recent times due to their wide-spread applications in carrying out chit-chat conversations with users and accomplishing various tasks as personal assistants. However, they still have some major limitations. One key limitation is that they are typically trained from pre-collected and manually-labeled data and/or written with handcrafted rules. Their knowledge bases (KBs) are pre-compiled by human experts and are fixed after deployment. Due to the huge amount of manual effort involved, they are difficult to scale and also tend to produce many errors ought to their limited ability to understand natural language and the limited knowledge in their KBs. Thus, when these systems are deployed, the level of user satisfaction is often low.

    In this tutorial, we introduce and discuss the research to give chatbots the ability to continuously and interactively learn new knowledge during conversation, i.e., learning on-the-job after deployment by themselves so that as the systems chat more and more with end-users, they become more and more knowledgeable and improve their performance over time. The first part of the tutorial focuses on introducing the paradigm of lifelong and continual learning and discuss various related problems and challenges in conversational AI applications. The second part presents the recent advancements in continual learning after model deployment, with a focus on continuous lexical and factual knowledge learning during chatting and learning to ground new language expressions via interactions with end-users. Finally, we conclude with a discussion on the scope for continual conversational skill learning and present some open challenges.

    See the original proposal for more details

    Date and Place

    Date: TBA
    Place: TBA

    Target Audience

    Researchers, graduate students, and practitioners who are interested in lifelong or continual learning, dialogue systems, and continual learning after deployment. It is particularly suitable for people who have been or will be invovled in building dialogue systems or chatbots.

    Prerequisite Knowledge

    Basic knowledge of machine learning and natural language processing.