Lecture 10

Question 1

3 principles of neural network processing

Answer 1

1. Distributed representations
2. Connectivity
3. Excitation/inhibition

Question 2

Why are representations distributed?

Answer 2

So that they are robust against noise, and damage, we have an increased capacity, and networks have an increased dynamic and flexible behaviour

Question 3

Connections between neurons (the synapses) are …

Answer 3

Plastic

Question 4

LTP

Answer 4

What fires together wires together

Question 5

Long term neuronal representation

Answer 5

Relatively strong synapses between the neurons of the ensemble

Question 6

What happens when you remember something?

Answer 6

Stimuli activate a part of the neuronal representation (cue), cue partially activates the patterns, then cued part of the ensemble of neurons activates the rest of the pattern

Question 7

Autoassociation

Answer 7

Local synaptic strengthening (within one layer) within one representation

Question 8

Heteroassociation

Answer 8

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

Question 9

General principles of conectivity

Answer 9

Divergence
Convergence
Point-to-point connectivity
(these patterns have a role in parallel processing)

Question 10

Divergence

Answer 10

Spread of information from one source cell to multiple target cells

Question 11

Convergence

Answer 11

Compression (combination) of information from multiple source cells onto 1 target cell

Question 12

Point-to-point (Topological)

Answer 12

Number of source and target neurons is the same. Copy of information from source to target layer

Question 13

Where in the hierarchy are connections often more plastic?

Answer 13

On higher levels

Question 14

Pattern completion

Answer 14

In memory tetreival: the cued part of the ensemble activates the rest of the pattern (most strongly interconnected neurons)

Question 15

Mechanisms through which inhibition is executed

Answer 15

Feedforward and feedback