BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:xCOMET\, Tower\, EuroLLM: Open & Multilingual LLMs for Europe
DTSTART:20250130T160000
DTEND:20250130T170000
DTSTAMP:20260405T091757Z
UID:278a7b07b78f2ccf925b0f7327cec95e54c57004615e0a2037d8370a
CATEGORIES:Conferences - Seminars
DESCRIPTION:André F. T. Martins\nAbstract: Today\, LLMs are Swiss knives 
 and MT one of their tools. Is this the end of MT research? In this talk\, 
 I argue that the connection between LLM and MT research is two-way. I pres
 ent some of our recent work advancing multilingual LLMs\, tools to estimat
 e their quality\, and how the two can be combined for test-time scaling.\n
 First\, I present xCOMET\, an open-source learned metric which integrates 
  sentence-level evaluation and error span detection\, exhibiting state-of
 -the-art performance across all types of meta-evaluation (sentence-level\,
  system-level\, and error span detection). Moreover\, it does so while hig
 hlighting and categorizing error spans\, thus enriching the quality assess
 ment.\nThen\, I present Tower\, a suite of open multilingual LLMs for tran
 slation-related tasks. Tower models are created through continued pretrain
 ing on a carefully curated multilingual mixture of monolingual and paralle
 l data. The combination of Tower with COMET reranking obtained the best re
 sults in 8 out of 11 language pairs in the WMT General Translation shared 
 task\, according to human evaluation.\nFinally\, I describe EuroLLM\, an o
 ngoing EU-made project whose goal is to train an open multilingual LLM fro
 m scratch using the European HPC infrastructure (EuroHPC). The last releas
 e (EuroLLM-9B) supports 35 languages\, including all 24 official EU langua
 ges\, and it achieves strong results in various benchmarks\, comparable or
  better than the best existing models of similar size.\n\nSpeaker:  Andr
 é F. T. Martins (PhD 2012\, Carnegie Mellon University and Instituto Supe
 rior Técnico) is an Associate Professor at Instituto Superior Técnico\, 
 University of Lisbon\, researcher at Instituto de Telecomunicações\, and
  the VP of AI Research at Unbabel. His research\, funded by a ERC Starting
  Grant (DeepSPIN) and Consolidator Grant (DECOLLAGE)\, among other grants\
 , include machine translation\, quality estimation\, structure and interpr
 etability in deep learning systems for NLP. His work has received several 
 paper awards at ACL conferences. He co-founded and co-organizes the Lisbon
  Machine Learning School (LxMLS)\, and he is a Fellow of the ELLIS society
  and co-director of the ELLIS Program in Natural Language Processing. He i
 s a member of the R&I advisory group of EuroHPC\, the European infrastruct
 ure for supercomputing.\n\n 
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/65202038736
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR