Two theoretical views

Question 1

Hierarchical model

Answer 1

Feedforward convergence of hierarhical RF properties. Starting from low level features to increasingly complex constellations, up to most specific (grandmother) cells.

Question 2

Dynamical assembly formation model

Answer 2

Cells encode specific low and higher level features, but objects are encoded by formation of dynamic assemblies, made possible through horizontal and feedback connections. Together they encode a particular object

Question 3

Cell types

Answer 3

Pontifical cells:
Highly specific cells, eg one cell for grandma, one for yellow Volkswagen etc.

Cardinal Cells:
In between the two, highly complex but a combination of cells assambled using different nodes.

Neural Assembly
Assembly of different neurons in mixed level features to assemble an object

Question 4

Assembly coding

Answer 4

Low and high feature information combines to form integated percept.

Needed for most objects, but there is also evidence for highly specific cells eg for faces.

Question 5

Combinatorial explosion

Answer 5

not possible to have one cell for every possible object, that’d lead to such explosion

Question 6

Ensemble coding theory

Answer 6

Solution to the combo explosion;
- Idea that grandmother is not represented by 1 neuron, but a set of neurons all coding for different features (faceness, hair, nose etc)
- Would also mean its more resistant to damage (loss of grandma cell…)
- More flexibility in novel objects
- Individual nodes participate in different ensembles

Question 7

Labeling essemblies

Answer 7

If same neurons are used, how do you know which belongs to which when presented with two objects.

Synchrony label theory
- Cross-correlation function of firing neurons reveals sync in potentials when A and B code for different object. Each node that belongs to an esembly fires in sync
- Only if two RFs stimulated by the same object

BUT doesn’t reflect figure ground segmentation regardless of cue