BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computational models of vision: From early vision to deep convolut
 ional neural networks - Prof. Felix A. Wichmann.    Neural Information Pro
 cessing Group\, Faculty of Science University of Tübingen
DTSTART:20170309T123000Z
DTEND:20170309T020000Z
UID:TALK71528@talks.cam.ac.uk
CONTACT:Dr. Milena Vurro
DESCRIPTION:Early visual processing has been studied extensively over the 
 last decades. From these studies a relatively standard model emerged of th
 e first steps in visual processing. However\, most implementations of the 
 standard model cannot take arbitrary images as input\, but only the typica
 l grating stimuli used in many of the early vision experiments.\nI will pr
 esent an image based early vision model implementing our knowledge about e
 arly visual processing including oriented spatial frequency channels\, div
 isive normalization and optimal decoding. The model explains the classical
  psychophysical data reasonably well\, matching the performance of the non
 -image based models for contrast detection\, contrast discrimination and o
 blique masking data. Leveraging the advantage of an image based model\, I 
 show how well our model performs for detecting Gabors masked by patches of
  natural scenes. Finally\, we observe that our model units are extremely s
 parsely activated: each natural image patch activates few units and each u
 nit is activated by few stimuli.\nIn computer vision recent and rapid adva
 nces in convolutional deep neural networks (DNNs) have resulted in image-b
 ased computational models of object recognition which\, for the first time
 \, rival human performance. However\, although DNNs have undoubtedly prove
 n their usefulness in computer vision\, their usefulness as models of huma
 n vision is not yet equally clear. On the one hand\, there is a growing nu
 mber of studies finding similarities between DNNs trained on object recogn
 ition to properties of the monkey or human visual system. At the same time
 \, however\, there are\, e.g.\, the well-known discrepancies as indicated 
 by so-called adversarial examples. Given our knowledge of early visual pro
 cessing\, a potential source for this difference may already originate fro
 m differences in the processing of low-level features. We performed object
  identification experiments with DNNs and human observers on exactly the s
 ame images under conditions favouring single-fixation\, purely feed-forwar
 d processing. Whilst we clearly find certain similarities\, we also find s
 trikingly non-human behaviour in DNNs\, as well as marked differences betw
 een different DNNs despite similar overall object recognition performance.
  I will discuss possible reasons for our findings in the light of our know
 ledge of early visual processing in human observers.\n
LOCATION:Lecture Theatre\, Medical Research Council Cognition and Brain Sc
 iences Unit
END:VEVENT
END:VCALENDAR