direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Modelle neuronaler Systeme

Ausgewählte Abstracts

Orientation-Specific Attention and Perceptual Learning in Recurrent Network Models of the Visual Cortex
Zitatschlüssel lars04
Autor Schwabe, L. and Obermayer, K.
Buchtitel Soc. Neurosci. Abstr. 30
Jahr 2004
Notiz CD-ROM
Zusammenfassung The representation of orientation information in the visual cortex is plastic as exemplified by phenomena like adaptation, perceptual learning and attention, but the induced changes are not the same: In perceptual learning no change [1] or even a \\\<i\\\>reduction\\\</i\\\> of activity for the learned orientation together with a shape change of the tuning function is reported whereas in attention only a multiplicative \\\<i\\\>increase\\\</i\\\> of activity is observed [2,3]. Does this imply that perceptual learning and attention subserve different functions? Here we hypothesize that both phenomena can be explained as an attempt to increase encoding accuracy around the learned/attended orientation.\\\<br\\\>We set up a recurrent network model with excitatory and inhibitory Hodgkin-Huxley-type neurons. Using a weighted average of the Fisher information to quantify encoding accuracy we determine the optimal changes for (i) the gain of the excitatory neurons, (ii) the values of the recurrent weights, and (iii) the strength of a modulatory input. We find that the changes induced when adapting only the recurrent weights resemble the findings which were reported for perceptual learning [1], including shifts of preferred orientations (POs), sharpening of the tuning curves, and modulation of the response amplitude. When only the gain and/or the modulatory inputs are adapted we find a multiplicative increase of the orientation tuning function similar to [2,3] which is strongest for neurons with POs 20 offset to the attended one, as well as subtractive shifts. We also demonstrate that an additive (hence not optimal) upward shift would also be consistent with [2,3]. Thus, experimental data is consistent with different kinds of mechanisms while for our hypothesis gain-modulation is required.\\\<br\\\>will show, how to distinguish between them experimentally.\\\<br\\\>\\\<br\\\>[1] Yang et al, J. Neurosci., 2004\\\<br\\\>[2] Treue et al, Nature, 1999\\\<br\\\>[3] McAdams et al., J. Neurosci., 1999
Download Bibtex Eintrag

Zusatzinformationen / Extras

Direktzugang:

Schnellnavigation zur Seite über Nummerneingabe