Events

CILVR Seminar: ATLAS: Adaptive Transfer Scaling Laws for Multilingual Pretraining, Finetuning, and Decoding the Curse of Multilinguality

Speaker: Shayne Longpre (MIT)

Location: 60 Fifth Avenue, Room 7th floor open space
Videoconference link: https://nyu.zoom.us/j/95718261115

Date: Wednesday, March 4, 2026

What if we need scaling and data mixing laws for non-English languages or combinations? ATLAS answers these questions with state-of-the-art multilingual scaling and mixing laws, guiding practitioners to make the most efficient and performant models for their objectives.

Bio: Shayne Longpre is a PhD candidate at MIT studying data-centric methods to build AI systems, as well as AI's socio-economic impact. He leads the Data Provenance Initiative, as well as initiatives to improve AI safety through audits and responsible disclosure. He has contributed to major model efforts, including BLOOM, Aya, and Flan-T5, and his work has earned Best Paper Awards at ACL 2024 and NAACL 2024, with coverage from outlets including The New York Times, Financial Times, and Washington Post.