BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Open-world learning across domains and modalities
DTSTART:20240229T163000
DTEND:20240229T183000
DTSTAMP:20260408T071301Z
UID:c37a6bc6cb3516f6153921652f8075cd5ecf23585adfc5a82869b768
CATEGORIES:Conferences - Seminars
DESCRIPTION:Shuo Wen\nEDIC candidacy exam\nExam president: Prof. Pascal Fr
 ossard\nThesis advisor: Prof. Maria Brbic\nCo-examiner: Prof. Robert West\
 n\nAbstract\nOpen-world learning is a paradigm in machine\nlearning that a
 ims to mimic human cognition by allowing models\nto continuously learn fro
 m new data and adapt to previously\nunseen classes or concepts. However\, 
 current approaches to openworld\nlearning rely on strong assumptions\, the
 reby restricting\ntheir applicability to a limited range of open-world sce
 narios.\nFor example\, many approaches assume that labeled and unlabeled\n
 data originate from the same domain\, which is not always true\nin real-wo
 rld scenarios. Overlooking crucial challenges such as\ndomain shift and mo
 dality shift can significantly affect model\nperformance when encountering
  data from different distributions\nor types. In this report\, we propose 
 to build an ideal open-world\nlearning model by dispelling all those assum
 ptions. Specifically\,\nthe model should be capable of perpetually learnin
 g new knowledge\nfrom diverse domains and modalities\, with applicability\
 nextending to various fields such as biomedicine.\n\nBackground papers\n\n
 	Towards Open World Recognition\n	Universal Domain Adaptation through Self
 -Supervision\n	Mapping Single-cell Data to Reference Atlases by Transfer L
 earning\n
LOCATION:BC 133 https://plan.epfl.ch/?room==BC%20133
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR