BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:IGLUE: A Benchmark for Transfer Learning across Modalities\, Tasks
 \, and Languages - Emanuele Bugliarello\, University of Copenhagen
DTSTART:20220310T110000Z
DTEND:20220310T120000Z
UID:TALK171323@talks.cam.ac.uk
CONTACT:Marinela Parovic
DESCRIPTION:Reliable evaluation benchmarks designed for replicability and 
 comprehensiveness have driven progress in machine learning. Due to the lac
 k of a multilingual benchmark\, however\, vision-and-language research has
  mostly focused on English language tasks. In this talk\, I will present t
 he Image-Grounded Language Understanding Evaluation benchmark that aims at
  filling this gap. IGLUE brings together—by both aggregating pre-existin
 g datasets and creating new ones—visual question answering\, cross-modal
  retrieval\, grounded reasoning\, and grounded entailment tasks across 20 
 diverse languages. Our benchmark enables the evaluation of multilingual mu
 ltimodal models for transfer learning\, not only in a zero-shot setting\, 
 but also in newly defined few-shot learning setups. Based on the evaluatio
 n of the available state-of-the-art models\, we find that translate-test t
 ransfer is superior to zero-shot transfer and that few-shot learning is ha
 rd to harness for many tasks. Moreover\, downstream performance is partial
 ly explained by the amount of available unlabelled textual data for pretra
 ining\, and only weakly by the typological distance of target–source lan
 guages. We hope to encourage future research efforts in this area by relea
 sing the benchmark to the community.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR