Task 4 - M&M

Question 1

How does competitive learning work ?

-How does it find a categorisation?

Answer 1

• is unsupervised
• network finds a categorisation for itself based on the similarity between input patterns and the number of output units available

Question 2

What do competitive learning networks learn?

Answer 2

learn to categorise input patterns into related sets, with one output unit firing for each set

Question 3

What happens when an input pattern is presented to a competitive learning network?

Answer 3

output units compete with each other to determine which has the largest response

Question 4

What happens with the connections to active and inactive input units to output units in competitive learning?

Answer 4

connections to winning output unit are strengthened and those from input units which were inactive are weakened

Question 5

How are the weights set in competitive learning?

Answer 5

weights are set by prior learning of network, not by an explicit external teacher

Question 6

What are the 3 phases of competitive learning?

Answer 6

excitation, competition and weight adjustment

Question 7

To which connections only are weight adjustments done in competitive learning?

Answer 7

only made to connections feeding into the winning output unit

Question 8

Which kind of learning rule does competitive learning use?

Answer 8

Uses a local learning rule

Question 9

In which way are competitive networks a feature of many real brain circuits ?

Answer 9

• Can remove redundancy-> allocates a single output neuron to represent a set of inputs which co-occur
• they can produce outputs for different input patterns which are less correlated with each other than the inputs were

Question 10

Which kind of associator is the auto- associator?

Answer 10

Form of pattern associator



Question 11

What is the aim of the auto- associator?

Answer 11

to reproduce the same pattern at output that was present at input

Question 12

What is the difference between pattern associator and autoassociator?

Answer 12

Auto-associator: output line of each unit is connected back to the dendrites of the other units -> recurrent connections

Question 13

What is the netinput in an autoassociator?

Answer 13

Netinput: external input and internal input, generated by feedback from other units within the autoassociator

Question 14

What are two features of auto-associators?

Answer 14

• Pattern completion

- Noise resistance

Question 15

What does the pattern associator learn?

Answer 15

learns to associate one stimulus with the other

Question 16

How does the training work for a pattern associator?

-What happens if learning is successful?

Answer 16

• training: pairs of patterns presented

- if learning successful: will recall one of the patterns at output when the other is presented at input

Question 17

What is the pattern associator able to do after training?

Answer 17

after training: can also respond to novel inputs, generalising from its experience with similar patterns

Question 18

How does pattern association take place?

Answer 18

pattern association takes place by modifying the strength of the connections between input units and output units

Question 19

What are properties of pattern associators? (6)

Answer 19

1. Generalisation: they generalise during recall
2. Fault tolerance: graceful degradation; small damage still leads to correct response
3. The importance of distributed representations for pattern associators: activity of all elements is used to encode a particular stimulus; generalisation and graceful degradation are only achieved if representations are distributed
4. Prototype extraction and noise removal: recognition of a prototype that has never been seen
5. Speed: fast due to parallel processing
6. Interference is not necessarily a bad thing: allows generalisation, noise reduction and prototype extraction; one reason why it is tolerated is that the ability to generalize between stimuli is more useful than 100% accurate memory of specific past events

Question 20

What is a recurrent network?

Answer 20

output line of each unit is connected back to the dendrites of the other units

Question 21

What is the delta rule about?

Answer 21

output unit can be corrected by increasing the weights of connections from units in the previous layer which provide a positive input to it and by decreasing the weights of the connections which provide a negative input

Question 22

What does saturation mean in context of Hebbian Learning?

Answer 22

If we continue to train the networks, it’s not possible that we can endlessly learn

Question 23

What is sparseness?

Answer 23

measure of the proportion of units which will be active in that area in response to an input

Question 24

What does sparse input enable?

Answer 24

sparse input enables an auto-associator to store more memories (e.g. dentate gyrus)

Question 25

What is a difference between the Delta and Hebbian learning rule?

Answer 25

• Delta rule: gradually smaller changes of the weights

- Hebb rule: only one learning trial may be required to establish the necessary connections

Question 26

Is it possible with Hebb’s law to build a computer model?

Answer 26

No, Hebb’s law is incomplete and inadequate to build a computer model

Question 27

What are the problems of Hebb’s law when it comes to using it as a computer model?

Answer 27

1. does not specify how much the connection between neurons should increase
2. does not specify how to compute the activity of the two neurons
3. there is nothing that ever allows the connection strength to decrease
4. does not specify the exact conditions under which the connections should strengthen

Question 28

What does Neo-Hebbian Learning consist of in its simplest form? (dynamical differential equations)

Answer 28

two sets of dynamical differential equations, one governing the activity change of an arbitrary network at a given instant in time and the other governing the weight changes of an arbitrary connection in the network at any instant in time

Question 29

What is an instar?

Answer 29

From the point of view of the neurode, it receives a large number of stimulus signals, coming from somewhere “outside” its boundaries
–> from its perspective, it is the center of an inwardly radiating collection of such signals

Question 30

What is an outstar?

Answer 30

• neurode sends its single output signal to a large number of other neurodes in the network
• when envisioning those outgoing signals as being more or less evenly distributed around the neurode, you can imagine an outwardly radiating star of output signals moving out from the neurode

Question 31

What is each neurode in terms of instars and outstars?

Answer 31

every neurode: both the center of an instar, receiving incoming stimuli from the outside and the center of an outstar, transmitting its output back to other instars or the outside world

Question 32

What do you look at in Neo-Hebbian learning?

Answer 32

you only look at change in connection strength

Question 33

What are problems of Neo- Hebbian learning?

Answer 33

• behavior of the outstar matches actual classical conditioning not in detail
• weights only increase in strength BUT a biological system cannot possibly increase without bonds

Question 34

How does the connection strength change in Differential Hebbian Learning?

Answer 34

the connection strength changes according to the change (difference) in the receiving neurode’s activation and the change in the incoming stimulus signal

Question 35

What are the problems of Differential Hebbian Learning?

Answer 35

• outstar exists only in a single, continuously changing moment of time called now -> only learn from stimuli that appear simultaneously BUT Pavlov’s experiment: learned more quickly when bell was rung before
• outstar: acquisition curve is linear; BUT in real life: S-shaped

Question 36

What do you look at in Differential Hebbian Learning?

Answer 36

you look at difference in change between connections

Question 37

What happens in Differential Hebbian Learning if either neurode has a constant activity level?

Answer 37

• if either neurode has a constant activity level, it has an activity change of zero -> no learning occurs
• the activity changes can be either positive or negative, the weight change can be either positive or negative

Question 38

Which two equations can be found in the Drive Reinforcement Theory (=DRT)?

Answer 38

• activity equation: describes the activation level of each neurode in the network
• weight change equation: describes how the connection strengths change during learning

Question 39

How can incoming signals contribute to the weighted sum in Drive Reinforcement Theory (=DRT)?

Answer 39

• each incoming signal must individually be greater than the threshold -> can contribute to the weighted sum
• an incoming signal less than or equal to the threshold value is considered a zero signal

Question 40

How do synaptic junctions work in DRT?

Answer 40

each synaptic junction is predetermined to be either an excitatory (positively weighted) or an inhibitory (negatively weighted) junction (always remains either positive or negative)

Question 41

Can weights be zero in DRT?

Answer 41

No