Invited Talk
in
Workshop: Wordplay: Reinforcement and Language Learning in Text-based Games

Hierarchical reinforcement learning for composite-task dialogues

Lihong Li

2018 Invited Talk
in
Workshop: Wordplay: Reinforcement and Language Learning in Text-based Games

Abstract

As in many complex text-based scenarios, a conversation can often be decomposed into multiple parts, each taking care of a subtopic or subtask that contributes to the success of the whole dialogue. An example is a travel assistant, which can converse with a user to deal with subtasks like hotel reservation, air ticket purchase, etc. In this talk, we will show how hierarchical deep reinforcement learning can be a useful framework for managing such "composite-task dialogues": (1) more efficient policy optimization with given subtasks; and (2) discovery of dialogue subtasks from corpus in an unsupervised way.

Speaker

Lihong Li

Lihong Li is an AI Research Scientist at Meta. He obtained a PhD degree in Computer Science from Rutgers University. After that, he has held research and applied science positions in Yahoo!, Microsoft, Google and Amazon. His main research interests are in large language models, recommendation systems, reinforcement learning, contextual bandits, and related areas in AI. His work has found applications in recommendation, advertising, Web search and conversational systems. He has won best paper awards at ICML, AISTATS and WSDM, as well as the 2023 Seoul Test of Time Award of the Web Conference. He regularly serves as area chair or senior program committee member at major AI/ML conferences such as AAAI, AISTATS, ICLR, ICML, IJCAI and NeurIPS. He has also served as associate, acting and guest editor for IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), Transactions on Machine Learning Research (TMLR), and Machine Learning Journal (MLJ).

Chat is not available.