A a b a c a neuron 1 the untangling hypothesis the

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: on 2 In IT, neurons maintain their stimulus preferences across position changes and the “response cloud” for the object (at the new position) remains on the correct side of the linear decision boundary. A A B A’ C A Neuron 1 The “Untangling Hypothesis” The “Untangling Hypothesis” The mean of the response cloud for this image The “Untangling Hypothesis” The mean of the response cloud for another view of the same object The “Untangling Hypothesis” All of the transformations (changes in position, size, pose, etc) of an object live on a “manifold” (a surface) in neural space The “Untangling Hypothesis” The manifold can be crumpled, flat (or anything, really) The “Untangling Hypothesis” Object recognition amounts to finding the “decision rule” that separates the manifolds for two different objects The “Untangling Hypothesis” ... V1 V2 V4 pIT IT At early stages of the visual system, the manifolds are thought to be crumpled (remember the “A” example); at later stages they are thought to be flat The “Untangling Hypothesis” “Implicit” information “Explicit” information Thus at early stages of the visual system,a linear decision rule will not separate the manifolds for two objects very well, whereas at higher stages it will work very well (again, remember the “A” example) The “Untangli...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online