Lecture 8

Question 1

Neural representation

Answer 1

Set of neurons that fire together and represent a concept/object/person.
No such thing as a single grandmother cell in this course

Question 2

Reasons for distributed representations

Answer 2

1. Robust against noise
2. Robust against damage
3. Increased capacity
4. Increases dynamic, flexible behaviour of networks

Question 3

How retrieval of memory works

Answer 3

A cue is presented which activates part of an inactive neural representation. Then the rest of the neural network is activated

Question 4

Autoassociation

Answer 4

Local synaptic strengthening within one representation

Question 5

Heteroassociation

Answer 5

Synaptic strengthening between different representations, different aspects of a stimulus, coded in different layers.

Question 6

Divergence

Answer 6

Spread of information from one source cell to multiple targets

Question 7

Convergence

Answer 7

Compression of information from multiple sources onto 1 target cell

Question 8

Topological

Answer 8

Number of source and target neurons is the same

Question 9

Divergence from source area to multiple target areas

Answer 9

Leading the same signal over parallel processing pathways to extract differential information.
E.g. info from retinal cells diverge over what and where pathway.

Question 10

Divergence from source area to one target area

Answer 10

Density of projection.
Spreading info over many target neurons to enable combination with any other stimuli. Like in associative memory.
High density projections more often in association areas.

Question 11

Convergence across modalities

Answer 11

On a larger scale, e.g. episodic memory system: hippocampus, a supramodal area, combines input from entire sensory field

Question 12

Convergence in general

Answer 12

Combination of simple features into progressively more complex aspects of the environment.

Question 13

Topological connectivity

Answer 13

Preservation of spatial organisation of environment, or source layer, in projection to target layer.
Some preservation of source layer topology at all levels of NS.
Relation to environmental topology lost at higher levels

Question 14

Differences in connectivity parameters across the processing hierarchy

Answer 14

• Connectivity pattern: lower levels more point to point, in association areas more divergence and often convergence
• Plasticity: connections often more plastic at higher levels of hierarchy

Question 15

Control of excitation by inhibition

Answer 15

• Fast inhibition through GABA-A receptors
• Feedforward and feeback mechanisms

Question 16

Feedforward inhibition

Answer 16

• Fast
• Proportional to activity in the source layer
• Allows network to handle input of variable strength

Question 17

Feedback inhibition

Answer 17

• Proportional to activity in target layer
• Limits excitation in time
• Limits pattern size
• Ensures that no associated patterns in the layer get activated
• Contributes to gamma oscillations