This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Cognitive Neuroscience Visual system: Ventral Route Hedge & Van Essen (2007) V2 vs. V4 Cartesian gratings NonCartesian gratings Contours For each cell: For each shape : Compute scaled response: shape lowest firing rate = 0.0 shape highest firing rate = 1.0 from 126 cells For each shape: For each area (V2, V4): Compute average scaled response Graph average for V2 vs V4 from 196 cells Conclusions V2: Gratings > Contours nonCartesian gratings > Cartesian gratings complex contours > simpler contours V4: Gratings >> Contours nonCartesian gratings > Cartesian gratings complex contours = simpler contours Work by Connor and Colleagues Present huge range (1000+) of twodimensional closed shapes See what features each cell responds to. V4 Sensitive to features of a single contour segment (part) curvature, orientation, relative position in object IT (inferotemporal) Sensitive to part configurations Specific combinations of contour segments NonLinear Response of an IT neuron to Part Configuration Response of an IT neuron sensitive to relative position of parts in object. Human Lateral Occipital Complex (LOC) Shape Perception LFD Local Features Deleted Hayworth & Biederman (2006) I identical C Complementary DE Different Exemplar P1 P2 P1 P2 Independent Variables: DeletionType (LFD, PD) P1/P2Relationship (I, C, DE) Task Is P2 the same exemplar as P1? Fusiform Gyrus pFs posterior Fusiform Gyrus: anterioventral portion of LOC Authors' Conclusion pFs region of LOC is encoding parts rather than vertices. Do you agree? Single cell recording evidence Visual categories: Specialized subsystems Neuropsychological evidence: Objects, Faces, Written Words Selective deficits affecting: Face recognition (prosopagnosia) Object recognition (visual agnosia) Reading (alexia) To account for this, we must assume that these categories must be represented with considerable independence in the brain Puce et al. (1996) fMRI: Passive viewing of alternating sequences of unfamiliar faces, unpronounceable letterstrings Faces RH>LH Letterstrings LH>RH 9/12 subjects showed distinct activation regions for letter and face processing within the inferior temporal lobe Are faces really a special kind of object? Have we evolved a face recognition module? With facespecific recognition processes? The alternative hypothesis: Specialized processes are used for all object types some object types recruit certain processes more than others By Francois and Jean Robert Ishai, et al. (1999) Tasks: 1) Passive viewing: single stimuli were presented at a rate of 2/sec Stimuli: photos of houses, faces, chairs or scrambled pix 2) Delayed match to sample: A sample image: 1.5 sec .5 sec delay 2 choice stimuli Ss indicated which of the choices corresponded to the sample Presentation: tasks were presented in 21 s blocks of the same type of stimuli, alternating with 21 s blocks of control scrambled stimuli Ishai et al. (1999) Conclusions: "Although our data could be interpreted as evidence for separate modules, we found that the responses to a category in ... regions that respond maximally to another category, were significant" "suggests that information most characteristic of ...a single category clusters together, resulting in a region that responds maximally to that a category and giving the appearance of a module. The debate continues..... Are faces really a special kind of object? First done on picture of Margaret Thatcher called the "Thatcher Illusion". 2i 1i 2 1 February 2002 issue of Discover Magazine (pg. 88), by Tanya Lombrozo Thatcher Illusion (new) On previous page, 2 is formed taking by 1 and turning the eyes and mouth upsidedown. 1i is just 1 inverted 2i is just 2 inverted 2 looks gross, but 2i does not. This shows that there's some kind of processing being performed on upright faces that's not being performed on inverted faces. This processing that is specific to upright faces is called configural or holistic. What this actually means is not well defined. Thatcher Illusion Interpreted as evidence for two different kinds of processing: Featural Upright and inverted faces Configural Upright faces only Thatcher Illusion (new) Configural (Holistic) Featural (Representionbyparts, Structural) X ? Martha Farah: Left Hemisphere Structural representations Objects, Letter Strings Right Hemisphere Holistic representations Objects, Faces Based on patterns of deficits in brain damage: Can have just alexia (LH) or just prosopagnosia (RH), but not object agnosia by itself. Alexia + prosopagnosia usually object agnosia Based on behavioral patterns in nonimpaired subjects: Subjects studied pictures of houses and faces, with owners' names. Then recognition of pictures of wholes, parts (e.g., Is this Larry's nose?) Accuracy: houses: wholes = parts (e.g. door) faces : wholes > parts (e.g. nose) RepetitionInduced Reductions in BOLD Signal Exp 1 Real = Nonsense Real > Nonsense Viewspecific Real Across exemplars of object Exp 2
Viewinvariant Real SizeInvariant RH Representation Seems like RH representation: is like a 2D picture (template) does not encode parts of objects is holistic, configural What the heck does that mean? How could there be a neural representation that is not based lowerlevel parts? Nobody has addressed this question. My Pet Theory We saw that 1D signal can be decomposed into a sum of sinusoids (e.g., sound, MEG and EEG signals) Similarly, a 2D signal (i.e., a picture) can be decomposed into sum of gratings. Proposal: RH representation is based on population code of gratings. LH contours, surfaces Parts are volumetric units (e.g., cylinders, spheres) RH repeating bars, swirls Parts are gratings ???? A B A,B: Same frequency, different phase I am not claiming these ideas are true. They are only proposals. However, you should understand the basic concept that a 2D picture can be decomposed into a weighted sum of gratings. If you don't understand this idea about phases, don't worry about it you won't be tested on it. Perhaps face analysis depends on phasespecific radial components. That is, perhaps there are neurons sensitive to pattern A, but not B. Perhaps configural analysis depends on the activation of neurons sensitive to pattern A, and such neurons are only activated by upright faces. Would explain sensitivity of configural analysis to inversion. Example of combining two Polar (non Cartesian) gratings to get a facelike image + = ...
View Full Document
This note was uploaded on 07/29/2008 for the course NEUROSCIEN 70 taught by Professor Whitney during the Spring '08 term at Johns Hopkins.
- Spring '08