Computational Neuroscience 1

Question 1

What is a model?

Answer 1

It’s an abstraction of a real system that incorporates mechanisms and relationships between key elements of it. To be useful, it should incorporate the minimum degree of complexity to enable predictions to be valid for the real counterpart.

Question 2

Why are models useful?

Answer 2

Models allow to focus on the fundamental mechanisms behind natural phenomena.

Question 3

Patient RB

Answer 3

Come back to

Question 4

Are people with hippocampal damage still capable of learning?

Answer 4

Yes, people with damage to the hippocampus still can learn new tasks like new motor skills, habits, and certain types of conditioning (i.e. emotional)

Question 5

What kind of memory does hippocampal damage most strongly affect?

Answer 5

Declarative

Question 6

What does the presence of anterograde with partial retrograde amnesia suggest?

Answer 6

That the hippocampus plays an important role in memory consolidation

Question 7

Classical condition

Answer 7

One of the most basic forms of associative memory (Pavlovian)

Question 8

Simplest neural networks

Answer 8

Feedforward network (unidirectional processing)
E.g. sensory systems, conditioning

Question 9

Two general types of neural network (real and artificial)

Answer 9

Feedforward

Recurrent

Question 10

Example of recurrent network

Answer 10

Memory storage

Question 11

McCulloch-Pits Neurone

Answer 11

First artificial neurone model proposed, still the core of all neural networks (1943)
DIAGRAM

Question 12

Transfer functions: sigmoidal vs linear

Answer 12

Sigmoidal transfer functions are more realistic, allow for staurable responses and more complex to use in calculations
Linear transfer functions are less realistic, unbounded responses and suitable for calculations of models

Question 13

What is learning in terms of a network?

Answer 13

Learning is the process of modifying the synaptic weights, so that the network achieves a specific behavioural goals

Question 14

Supervised learning

Answer 14

In this case there is a ‘teacher’ that tells the network whether the response given to the test inputs is correct or not. This feedback is then used by the network to decide how to modify the weights. Useful for learning tasks whose success is determined by the environment.

Question 15

Unsupervised learning

Answer 15

There is not ‘teacher’ or external feedback. The rules for synaptic weight modification are already embedded in the network definition. Useful for associative memory storage and retrieval.