Our weekly SRI Seminar Series welcomes Marzieh Fadaee, a senior research scientist at Cohere For AI, a non-profit research lab that seeks to solve complex machine learning problems and create more points of entry into machine learning research.
Fadaee’s work is broadly interested in all aspects of natural language understanding, particularly in multilingual learning, data-conscious learning, robust and scalable models, compositionality, and interpretability. In this talk, she will explore some of the current challenges in the latest developments of large language models (LLMs), including improving their multilingual capabilities, fine-tuning, support for low-resource languages, and mitigating bias.
“Mastering language understanding with AI: How multilingualism shapes LLMs”
Large language models (LLMs) have emerged as transformative tools in natural language processing (NLP), playing a pivotal role in various domains. This presentation offers an in-depth exploration of the evolution of language models, tracing their origins to machine translation and multilingualism. It dissects the critical advancements in machine translation that laid the groundwork for developing LLMs.
Since their introduction, LLMs have consistently performed remarkably in linguistic tasks as well as domains like law, medicine, and education. Their versatility across different fields highlights their significant and wide-ranging contributions. Despite their remarkable capabilities, LLMs face some challenges in fully realizing their potential. Coming full circle, some of the challenges of today’s LLMs stem from the multilingual capabilities of these models. The next big advancement for these models is improving their multilingual capabilities (e.g., addressing issues related to fine-tuning specific languages, handling low-resource languages, and mitigating biases). Evaluating how LLMs perform across diverse languages and cultural contexts is imperative for harnessing their full potential in a globalized world.
Michael Tomasello, Constructing a Language: A Usage-Based Theory of Language Acquisition. Harvard University Press, 2005.
Jean Kaddour, Joshua Harris, Maximilian Mozes, Herbie Bradley, Roberta Raileanu, Robert McHardy, “Challenges and Applications of Large Language Models,” arXiv, 19 July 2023.
Marzieh Fadaee is a senior research scientist at Cohere For AI. Her work is broadly interested in all aspects of natural language understanding, and particularly in multilingual learning, data-conscious learning, robust and scalable models, compositionality, and interpretability.
Previously, Fadaee was the NLP/ML research lead at Zeta Alpha Vector working on smarter ways to discover and organize knowledge in AI. Fadaee received her PhD from the University of Amsterdam, where she worked with the Language Technology Lab, a research group focusing on information access from natural language data. She previously received her BSc in computer engineering from Sharif University and MSc in artificial intelligence from the University of Tehran.
To register for the event, visit the official event page.
The SRI Seminar Series brings together the Schwartz Reisman community and beyond for a robust exchange of ideas that advance scholarship at the intersection of technology and society. Seminars are led by a leading or emerging scholar and feature extensive discussion.
Each week, a featured speaker will present for 45 minutes, followed by an open discussion. Registered attendees will be emailed a Zoom link before the event begins. The event will be recorded and posted online.