Task 4 Episode 4

Question 1

Competetive learning

Answer 1

A form of unsupervised learning in artificial neural networks, in which nodes compete for the right to respond

Question 2

Hebb rule

Answer 2

• Rule that states that the connections between two neurons will be strengthened if the neurons fire simultaneously
• If there is activity on input axon j when neuron i is active, then the strength of the connection (wij) between axon j and dendritre i is increased

Question 3

Saturation

Answer 3

Weights between nodes in an artificial neural network only increase as they fire together, which can leave the network overactive, because at some point every neuron starts to fire (Solved by condition, that the sum of all input to one neuron cannot be more than 1)

Question 4

Interference

Answer 4

Weights between nodes in a neural network will interfer with learning, because one weight can only store a limited amount of information

Question 5

Auto-associator

Answer 5

• Neural networks that reproduces the input as output, and can remove noise from the input (Noise removal), and complete missing signals from the input (Pattern completion)
• Recurrent connections
• Supervised learning

Question 6

Pattern-associator

Answer 6

• A pattern associator learns associations between input patterns and output patterns, and can link input stimuli that are presented together
• Comparable to classical conditioning
• Noise tolerance
• Fault tolerance
• Unsupervised

Question 7

Recurrent network

Answer 7

Connections between nodes that can exhibit temporal dynamic behaviour by feeding the output back as its input

Question 8

Hippocampus

Answer 8

Structure in the brain that is responsible for forming new episodic memories.

Question 9

Drive reinforcement theory

Answer 9

Reinforcement learning is an autonomous, self-teaching system that essentially learns by trial and error

Question 10

Dentate gyrus (DG)

Answer 10

A brain region of the hippocampus which forms inhibitory connections and in which competitive learning takes place.

Question 11

CA3

Answer 11

A brain region of the hippocampus which has a recurrent connection and in which competitive learning takes place

Question 12

CA1

Answer 12

A brain region of the hippocampus that acts as a pattern associator

Question 13

Constraint satisfaction

Answer 13

Network trys to satisfy all constraints (limitations), therefore each neuron influences the state of all other neurons and each input is a constraint (Not all constraints can be satisfied)

Question 14

Bias

Answer 14

• Additional units that bias the activity of the units in a layer
• Represents negative threshold
• Allows to have specific threshold for each neuron by changing weight from bias to neuron

Question 15

Fault tolerance

Answer 15

• System is resistant to small damage and only has small decrease in functioning as damage increases

Question 16

Content-addressible memory

Answer 16

• Cued recall

- Part of information available –> Produce whole content