Details
TO ATTEND THE EVENT PLEASE REGISTER HERE.
Join us for an exciting evening of discussion and networking at the AI Meetup on June 6, 2024. This event will feature insightful talks from leading experts in the field of artificial intelligence and natural language processing.
Agenda:
- 5:30 pm - Arrival & Networking
- 6:00 pm - "Is a large language model a (new) model for dialogue?" by Prof. Dr. Milica Gašić, Heinrich-Heine-University
- 6:45 pm - Break
- 7:00 pm - "Should we use Large Language Models (LLMs) as evaluators?" by Dimitra Gkatzia, Associate Professor, Edinburgh Napier University
- 7:45 pm - Break
- 8:00 pm - "Our Journey with Hotel Highlights" by Srinivas Kamath, Data Scientist, trivago
- 8:20 pm - Dinner and networking
Talk 1: Is a large language model a (new) model for dialogue?
Abstract: ChatGPT caused a lot of excitement when it appeared in November 2022. Since then we saw a lot of expressions of concern, but also newer and better models. But what is behind the large language models’ exceptional performance? What is new about this technology? And what does that mean for dialogue modelling? These are some of the questions that I’ll try to address in this talk.
Bio: Prof. Dr. Milica Gašić is a Professor in Dialog Systems and Machine Learning at Heinrich Heine University. Her research focuses on fundamental questions of human-computer dialogue modelling and lie in the intersection of Natural Language Processing and Machine Learning. She is a recipient of the European Research Council Starting Grant and the Alexander von Humboldt Sofja Kovalevskaja Award. Prof. Gašić is the vice-president of SIGDIAL (Special Interest Group on Discourse and Dialogue), a member of the International Scientific Advisory Board of DFKI (Deutsches Forschungszentrum für Künstliche Intelligenz), a member of ACL (Association for Computational Linguistics), a member of ELLIS (European Lab for Learning & Intelligent Systems) and a senior member of IEEE (Institute of Electrical and Electronics Engineers).
Talk 2: Should we use Large Language Models (LLMs) as evaluators?
Abstract: Large Language Models (LLMs) have changed the landscape of research in natural language processing, enabling unprecedented advancements in tasks ranging from text generation to understanding. Recently, there have been many efforts to use LLMs to evaluate automatically generated text instead of human evaluations, which the research community agrees is a more robust and reliable evaluation practice. Therefore, the question of employing LLMs as evaluators in various contexts remains contentious. In this talk, I will first discuss the current norms of evaluation of automatically generated text, focusing on popular tasks such as dialogue, summarisation (including data-to-text) and translation and elaborate on issues related to human evaluation and automatic evaluation. Then, the talk will review the main approaches to evaluation with LLMs as judges and will shed some light on the multifaceted considerations surrounding the utilisation of LLMs as evaluators.
Bio: Dimitra Gkatzia is an Associate Professor at the School of Computing at Edinburgh Napier University. She is interested in the evaluation of Natural Language Generation systems, and data-driven Natural Language Generation (NLG) for low-resource domains/languages. More recently, Dimitra has been interested in exploring privacy issues in LLMs and exploring how we can use them responsibly. Since 2021, Dimitra is the co-lead of the SICSA (Scottish Informatics and Computer Science Alliance) AI Theme. Between 2016-2020, Dimitra served as an elected member of the Steering Board of the Special Interest Group in Natural Language. Generation (SIGGEN). At Edinburgh Napier, Dimitra is currently leading the Natural Language Processing Group.
Talk 3: Our Journey with Hotel Highlights
Abstract: Hotel Highlights is our latest Generative AI feature that leverages a LLM to create unique summaries of our accommodations. In this talk we give a peek into the user and business driven decisions that we took to design this feature. We focus on the creation of this feature, the challenges that we faced, and the learnings that we gained. We discuss the different components that surround a LLM based feature and the significance of metadata to augment the creation of generated content.
BIO: Srinivas Kamath is a Data Scientist at trivago, who has for the last 5 years focused on organising, interpreting, and visualising accommodation content data. He has developed several key algorithms within trivago, such as classifying accommodation types, drawing on his many years of domain expertise. Recently, he has worked more extensively on applied natural language processing problems, focusing on aspects such classification and summarisation.
