BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Question Answering in Realistic Visual Environments: Challenges an
 d Approaches - Catalina Cangea
DTSTART:20191029T130000Z
DTEND:20191029T140000Z
UID:TALK131932@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:The Embodied Question Answering (EQA) and Interactive Question
  Answering (IQA) tasks were recently introduced as a means to study the ca
 pabilities of agents in rich\, realistic 3D environments\, requiring both 
 navigation and reasoning to achieve success. Each of these skills typicall
 y needs a different approach\, which should nevertheless be smoothly integ
 rated with the rest of the system leveraged by the agent. However\, initia
 l approaches either suffer from potentially weaker performance than when u
 sing a language-only model or are preceded by additional hand-engineered s
 teps. This talk will provide an overview of the existing work on this thre
 ad and describe in more detail our recent study (which I started while bei
 ng a research intern at Mila last summer) published at BMVC 2019 and to be
  presented as a spotlight talk at the ViGIL NeurIPS 2019 workshop\, "Video
 NavQA: Bridging the Gap between Visual and Embodied Question Answering":ht
 tps://arxiv.org/abs/1908.04950. Slides are available "here":https://drive.
 google.com/file/d/10J1xw3L2WOqE5n7xX5uISlXZwy_r9ElS/.
LOCATION:SS03\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR