Group Meeting Schedule

Fall, 2013

Potential papers for discussion are provided here (accessible only within Psych and CNS).
Send email to Eero if you have more suggestions!

Date Presenter Topic
Tue, 3 Sep, 2:00pm Jesus Malo
U. Valencia
Fitting multi-stage models using MAximum Differentiation (MAD)
Tue, 10 Sep, 2:00pm Neil Rabinowitz
LCV
Discussion: "The impact on midlevel vision of statistically optimal divisive normalization in V1",
Coen-Cagli & Schwartz, JOV 2013
Wed, 18 Sep, 3:30pm Johannes Ballé
LCV
Convolutional feed-forward networks, ICA, and GSM: What's the objective?
Wed, 2 Oct, 2:00pm Joint meeting with Ma lab Research overview
Wed, 9 Oct, 2:00pm Matt Zeiler
Fergus lab
Deep networks for visual recognition
Wed, 29 Oct, 2:00pm Elad Ganmor
LCV
Efficient coding: what is it, and how strong is the evidence?
Wed, 6 Nov, 2:00pm Michael Okun
UCL
Structure of spontaneous activity in rodent sensory cortex
Wed, 13 Nov, 2:00pm Neil Rabinowitz
LCV
Neurons are not stationary. Let's stop pretending they are.
Wed, 27 Nov, 2:00pm Sebastiaan van opHeusden
CNS
Zipf's law: a signature of criticality or something completely different?
On different trials identical sensory stimuli elicit distinct neuronal responses. Furthermore, even in the absence of sensory stimuli the cortex produces unpredictable but structured spontaneous activity that in many ways resembles sensory responses. According to a fascinating recent hypothesis, these are hallmarks of sampling-based representation used by the cortex to perform sensory inference computations (Berkes et al., Science 2011). In the first part of the talk I will present a different interpretation of the currently existing evidence of sampling-based representation. The central concept in our alternative explanation is population rate dynamics, i.e., the propensity of nearby cortical neurons to change their firing rate in a coordinated manner. In the second part I will describe how the spiking of individual neurons correlates with the population rate signal. We find that during both spontaneous and evoked activity the magnitude of this correlation is highly heterogeneous even across neighboring neurons. Even though response normalization is a canonical neural computation that happens at every stage of the visual pathway (Carandini & Heeger, Nat. Rev. Neurosci. 12), not many vision models explicitly use this modular structure in cascade. Exceptions include the two-stage model in (Simoncelli & Heeger, Vis. Res. 98), or the recent two-stage model in (Kay, Winawer & Wandell, PLoS Comp. Biol. 13). One possible reason is because there are too many parameters to fit. Here we show how to select the parameters of this kind of cascade models using simple psychophysics based on Maximum Differentiation (Wang & Simoncelli, JoV 08). As an illustration of the meaningfulness and generality of the parameters, we show that the obtained two-stage model beats state-of-the-art full-reference metrics for image quality.

Previous Semester Schedules:   Summer 2013 | Spring 2013 | Fall 2012 | Summer 2012 | Spring 2012 | Fall 2011 | Spring 2011 | Fall 2010 | Summer 2010 | Spring 2010 | Fall 2009 | Summer 2009 | Spring 2009 | Fall 2008 | Spring 2008 | Fall 2007 | Spring 2007 | Fall 2006 | Summer 2006 | Spring 2006 | Fall 2005 | Summer 2005 | Spring 2005 | Fall 2004 | Summer 2004 | Spring 2004 | Fall 2003 | Summer 2003 | Spring 2003 | Summer 2002 | Spring 2002 | Fall 2001 | Summer 2001 | Spring 2001