BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Second order statistics in deep learning
DTSTART:20170626T093000
DTEND:20170626T113000
DTSTAMP:20260407T105902Z
UID:ed1afa298f75c18b9cf7576751958398b2638adffce300c930928aae
CATEGORIES:Conferences - Seminars
DESCRIPTION:Kaicheng Yu\nEDIC candidacy exam\nExam president: Prof. Martin
  Jaggi\nThesis advisor: Prof. Pascal Fua\nThesis co-advisor: Dr. Mathieu S
 alzmann\nCo-examiner: Prof. Robert West\n\nAbstract\nConvolutional neural 
 networks (CNNs) have been shown effective in many visual recognition tasks
  recently. However\, a CNN is formulated by weighted summation which limit
 s the network to explicitly learn second-order statistics.\nI will study t
 hree related works in this report\, region covariance descriptors in pedes
 trian detection\, a general back-propagation algorithm to support matrix l
 evel gradient descent and a recent state-of-the-art deep network architect
 ure DenseNet and demonstrate the relevance to our future research goal\, w
 hich is to introduce second-order statistics in deep networks. In addition
 \, I will briefly discuss our current research\, global covariance descrip
 tor in CNNs\, and our future research plan.\n\nBackground papers\nPedestri
 an Detection via Classification on Riemannian Manifolds\, IEEE paper.\nMat
 rix Backpropagation for Deep Networks with Structured Layers Shorter versi
 on\, Ionescu C.\, et al.\nDensely Connected Convolutional Networks\, Huang
  G.\, et al.
LOCATION:BC 329 https://plan.epfl.ch/?room==BC%20329
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR