This preview shows page 1. Sign up to view the full content.
Unformatted text preview: ng” Hypothesis - as signals pass through the “what” pathway, the
manifolds for different object “untangle”
Retina V1 IT In other words, the purpose of the “what” pathway is to untangle manifolds
(for invariant object recognition) pIT
IT LGN *How* are object manifolds untangled?
retina Selectivity Receptive
field size retina V1 V4 IT Hypothesis: in each visual area, the neurons become a little more selective for
conjunctions of simpler features and and a little more tolerant for changes in the
position and size of those features
Until signals reach IT where the neurons become tolerant for changes in the position and
size of objects!
As for changes in pose, well, we donʼt know ... LGN neurons combine to form orientation-tuned V1 simple cells
(selectivity) “LN model” Hubel & Wiesel, 1962 Simple cells combine to form position/polarity invariant complex cells
“LN-LN cascade” Hubel & Wiesel, 1962 Adelson & Bergen, 1985 V1 neurons combine to form V2 corner detection neurons ARTICLES a 2 b V2 1.0 1 0.5
0 0 –0.5 –1
–2 d V2 24 spikes per s 0 –1.0 2
V2 11 spikes per s –1 e 2 0 1
V2 1 y (°) 1
0 0 –1
16 spikes per s –2 14 spikes per s c2 Figure
18 spikes per s
Anzai, Peng, Van Essen 2007)
32 spikes per s f V2 To build up an IT population representation that is selective for objects
invariant of the details (like position), the visual system implements a little
selectivity and invariance in each visual area...
View Full Document
This note was uploaded on 09/08/2013 for the course BBB 217 taught by Professor Nicolerust during the Spring '12 term at UPenn.
- Spring '12