A Profª Thamar Solorio, da Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) em Abu Dhabi, estará visitando o INF-UFRGS no dia 19/02 (segunda-feira). Ela irá apresentar uma palestra sobre Visual Question Answering. Além disso, uma representante da universidade irá discutir oportunidades de intercâmbio para alunos de graduação e pós nas áreas de aprendizado de máquina, processamento de linguagem natural, visão computacional e robótica. Os alunos selecionados ganham bolsa de estudos, passagens e outros benefícios.
A palestra será às 12:45 no Auditório 0.
Título: Exploring the Limits of Textual Information and Open Models for Video Question Answering Tasks
Abstract: Video Question Answering (vQA) deals with the problem of answering questions that can be resolved by watching a video, usually a short one, of around one minute or so. The current state of the art (SOTA) solutions to vQA consist of a complex arrangement of large vision and language models, with usually expensive end to end training involved and/or relying on APIs of closed models. Yet, there is still a huge performance gap from these sophisticated solutions compared to human performance. My group has been exploring the adaptation of SOTA solutions to vQA problems with limited open domain models and lower compute needs. In particular we have been focusing on generating richer textual representations of the video frames. During this talk I will present our recent results and lessons learned on these efforts.
Bio: Thamar Solorio is a Professor in the department of NLP at MBZUAI. She is also a tenured professor of Computer Science at the University of Houston. She is the director and founder of the RiTUAL Lab. Her research interests include multilinguality and low resource NLP, as well as information extraction, and more recently, language and vision problems. She is the recipient of an National Science Foundation (NSF) CAREER award for her work on authorship attribution, and recipient of the 2014 Emerging Leader ABIE Award in Honor of Denice Denton. She just completed two terms as elected board member of the North American Chapter of the Association of Computational Linguistics (NAACL) and was PC co-chair for NAACL 2019. She is an Editor in Chief for the ACL Rolling Review (ARR) initiative and member of the advisory board for ARR. Her research has been funded by NSF and ADOBE.