Representation of a 3D object:
a pointer to the real thing, not an internal replica or description
- Philosophical arguments:
- Internal reconstruction is unnecessary/insufficient for useful representation
- Putnam and Millikan on the meaning being not in the head.
- Locke on covariation ("conformity") and Shepard on 2nd-order isomorphism between the representational system and the world.
- Psychophysical evidence:
- O'Regan on the world as an external memory.
- Poor representation of geometrical structure (1st-order isomorphism is out)
- Blackmore et al. on poor representation of absolute structure of scenes: Richness of the immediate visual experience of a scene, contrasted with poverty for detail of the memory trace generated by the same scene, supports the idea of representation being a pointer to reality.
- Good representation of similarity patterns (2nd-order isomorphism is in)
- Edelman on veridical representation of relative structure (similarity relationships) of collections of 3D shapes. The subjects' veridical perception of the parametric structure of shape space suggests that the representations refer to the relationships among shapes (as per Shepard's idea of 2nd-order isomorphism) rather than to the geometry of each shape.
- Computational considerations:
- 2nd-order isomorphism can be easily supported by a Chorus of coarsely tuned classifiers
Under this approach, the features used for representing 3D objects are holistic (based on similarities to entire reference objects rather than on structures composed of parts that require binding) and pointer-like (based on the principle of referring to the world rather than reconstructing its replica internally).
Go on to the next section.