Yun-Nung (Vivian) Chen
Associate Professor / National Taiwan University
Robustness, Scalability, and Practicality of Conversational AI
Abstract: Even conversational systems have attracted a lot of attention recently, there are many remaining challenges to be resolved. This talk presents three different dimensions for improvement: 1) Robustness — how to deal with speech recognition errors for better language understanding performance, 2) Scalability — how to better utilize the limited data, and 3) Practicality — how to naturally perform recommendation in a conversational manner. All directions enhance the usefulness of conversational systems, showing the potential of guiding future research areas
Bio: Yun-Nung (Vivian) Chen is currently an associate professor in the Department of Computer Science & Information Engineering at National Taiwan University. She earned her Ph.D. degree from Carnegie Mellon University, where her research interests focus on spoken dialogue systems and natural language processing. She was recognized as the Taiwan Outstanding Young Women in Science and received Google Faculty Research Awards, Amazon AWS Machine Learning Research Awards, MOST Young Scholar Fellowship, and FAOS Young Scholar Innovation Award. Her team was selected to participate in the first Alexa Prize TaskBot Challenge in 2021. Prior to joining National Taiwan University, she worked in the Deep Learning Technology Center at Microsoft Research Redmond.
Staff Research Scientist / DeepMind
On opportunities and challenges on communicating using Large Language Models
Abstract: From science fiction to Turing’s seminal work on AI, language and communication have been among the central components of intelligent agents. Towards that dream, the new-generation of large language models (LLMs) have recently given rise to a new set of impressive capabilities, from generating human-like text to engaging in simple, few-turn conversations. So, how close do LLMs bring us to being able to interact with such intelligent agents during our lifetime? In this talk, I will review key recent developments on LLMs by the community and I will discuss these in the context of advancing communication research. At the same time, I will also highlight challenges of current models in producing goal-driven, safe and factual dialogues. Capitalizing on their strengths and addressing their weaknesses might allow us to unlock LLMs full potential in responsibly interacting with us, humans, about different aspects of our lives.
Bio: Angeliki Lazaridou is a Staff Research Scientist at DeepMind. She received a PhD in Brain and Cognitive Sciences from the University of Trento. Her PhD initially focused on developing neural network models and techniques for teaching agents language in grounded environments. However, one day in late 2015, while walking towards the lab she realized that interaction and communication should play a key role in this learning 💡. This was the beginning of her work in deep learning and multi-agent communication. In the following years, she looked at this fascinating problem from many different angles: how to make this learning more realistic or how to extend findings from cooperative to self-agents and even how to make this communication resemble more natural language. Currently, she spends most of her time thinking and working on how to best make language models be in sync with the complex and ever-evolving world.
Professor / University of British Columbia
Unlimited discourse structures in the era of distant supervision, pre-trained language models and autoencoders
Abstract: Historically, discourse processing relies on human annotated corpora that are very small and lack diversity, often leading to overfitting, poor performance in domain transfer, and minimal success of modern deep-learning solutions. So, wouldn’t it be great if we could generate an unlimited amount of discourse structures for both monologues and dialogues, across genres, without involving human annotation? In this talk, I will present some preliminary results on possible strategies to achieve this goal: by either leveraging natural text annotations (like sentiment and summaries), by extracting discourse information from pre-trained and fine-tuned language models, or by inducing discourse trees from task-agnostic autoencoding learning objectives. Besides the many remaining challenges and open issues, I will discuss the potential of these novel approaches not only to boost the performance of discourse parsers (NLU) and text planners (NLG), but also lead to more explanatory and useful data-driven theories of discourse.
Bio: Giuseppe Carenini is a Professor in Computer Science and Director of the Master in Data Science at UBC (Vancouver, Canada). His work on natural language processing and information visualization to support decision making has been published in over 140 peer-reviewed papers (including best paper at UMAP-14 and ACM-TiiS-14). Dr. Carenini was the area chair for many conferences including recently for ACL’21 in “Natural language Generation”, as well as Senior Area Chair for NAACL’21 in “Discourse and Pragmatics”. Dr. Carenini was also the Program Co-Chair for IUI 2015 and for SigDial 2016. In 2011, he published a co-authored book on “Methods for Mining and Summarizing Text Conversations”. In his work, Dr. Carenini has also extensively collaborated with industrial partners, including Microsoft and IBM. He was awarded a Google Research Award in 2007 and a Yahoo Faculty Research Award in 2016.