Cosyne 2007 Workshops
February 26-27, 2007
The Canyons, Utah
Workshop Title
Functional Requirements of a Visual Theory
Organizer(s)
David Arathorn: dwa@giclab.com
Bruno Olshausen (UC Berkeley): baolshausen@berkeley.edu
Jim DiCarlo: dicarlo@mit.edu
Abstract
The objective of this workshop is to gather a group of investigators who are interested in building theories of the visual system and want to identify hard but important questions in vision are not being commonly asked, or widely discussed, and even less pursued experimentally or theoretically.
If one attempts to apply the existing body of neuronal hypotheses on the mechanics of vision to real world vision problems one immediately discovers there are substantial areas essentially unaddresssed. For example, if a student asks how does the visual system supply its owner with a navigable three dimensional model of a rock outcropping viewed monocularly, one would be hard pressed to find either a computational model or any compelling evidence of which visual area or areas host such a computation. More curiously, there is not much attention paid in the literature to the fact that this is an important unanswered question. Instead, texts in which facts and theories about the visual system are compiled tend to be data-driven: even important questions for which there are no experimental results are usually absent. One of the consequences of being data-driven is that the efforts in the field tend to get blown into overpopulated corners by the prevailing methodological winds.
The inspiration here is taken from common engineering practice: if you are going to build something, you first compile a functional requirements document which should be a reasonably complete description of what the something is required to do. This is an antidote to the natural tendency to solve the easy problems first, only to find that the solutions one has adopted at the beginning preclude solutions for the hard problems later on.
This will be a `working' workshop. There will be a limited number of opening presentations on important, underaddressed capabilities of the visual system, current theoretical cul-de-sacs, and other issues which should be under consideration in the construction of any working theory of vision. The majority of the time attendees will be expected to participate in compiling an outline of visual functions, the importance, the level of current understanding, who is doing experimental or theoretical work relevant to that area. When this document is later cleaned up it can serve as a map for students or others interested in informing themselves or working in particular areas. It should also hopefully serve to guide funding agencies to direct means to un-addressed, or under-addressed, areas of research. If the effort comes to be judged as useful, this document can be kept up to date by regular re-occurrences of this workshop.
Panelists
David Arathorn (GIC)
Bruno Olshausen (UC Berkeley)
Jim DiCarlo (MIT)
Charles Anderson (Washington University)
David Field (Cornell)
Yann LeCun (NYU)
Michael Lewicki (CMU)
Fei-Fei Li (Princeton)
Andrew Ng (Stanford)
Aude Oliva (MIT)
Thomas Serre (MIT)
Antonio Torralba (MIT)
Alan Yuille (UCLA) - tentative
