BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Deep Structured Representation Learning for Visual Recognition
DTSTART:20180703T100000
DTEND:20180703T120000
DTSTAMP:20260407T164205Z
UID:6206e017322e86574fc12a73723218fcdbb6d39989b4d4d033abcd22
CATEGORIES:Conferences - Seminars
DESCRIPTION:Krishna Kanth Nakka\nEDIC candidacy exam\nExam president: Prof
 . Sabine Süsstrunk\nThesis advisor: Prof. Pascal Fua\nThesis co-advisor: 
 Dr Matthieu Salzmann\nCo-examiner: Prof. Pierre Dillenbourg\n\nAbstract\nS
 tructured representations\, such as Bags of Words\, VLAD and Fisher Vector
 s\, have proven highly successful to tackle complex visual recognition tas
 ks. As such\, they have been incorporated into deep architectures. Our goa
 l is to develop deep attentional architectures that jointly learn an inter
 pretable semantic codebook and a structured representation of the input im
 age to better understand of the way deep networks perform visual recogniti
 on tasks. Our framework is designed to understand how networks make predic
 tions\, as well as when and why they make errors. We leverage the semantic
  codebooks to detect malicious inputs in case of adversarial attacks and p
 ropose a defense system to tackle the same. We first show the effectivenes
 s of structured representations for the task of large scale image retrieva
 l. We then present a visualization technique using generator networks to i
 nterpret the hidden neurons of a deep network  and finally discuss a stat
 e-of-the-art method on visual attention system that focus on discriminativ
 e regions of the image. We conclude with the current state of results and 
 directions for future research.\n\nBackground papers\nAggregating local de
 scriptors into compact codes\, PAMI 2012 \nSynthesizing the preferred inp
 uts for neurons in neural networks via deep generator networks\, NIPS 2016
 \nAttentional Pooling for Action Recognition\, NIPS 2017.
LOCATION:BC 329 https://plan.epfl.ch/?room==BC%20329
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR