Week/Topic 7: Connectionism

Question 1

Connectionist networks

Answer 1

are a form of
computation that is loosely inspired by the
way the brain works

Question 2

Connectionist networks are (roughly) the
same as:

Answer 2

Neural Networks
Parallel Distributed Processing (PDP)

Question 3

Connectionism

Answer 3

is the view that such
networks are useful for understanding
how information is processed in the mind

Question 4

Perceptron Convergence Rule

Answer 4

A learning algorithm for perceptron’s (single-unit networks). It changes a perceptron’s threshold and weights as a function of the difference between the unit’s actual and intended output.

Question 5

Hebbian Rule

Answer 5

a model of how neurons learn and form connections in the brain

Question 6

Difference between Hebbian Rule and Perceptron-Convergence Rule

Answer 6

Hebbian learning rule – It identifies, how to modify the weights of nodes of a network. Perceptron learning rule – Network starts its learning by assigning a random value to each weight

Question 7

Who came up with the Perceptron-Convergence rule, and when?

Answer 7

While the perceptron algorithm was invented by Rosenblatt, this theorem and its proof were from NYU mathematician Albert B. J. Novikoff in 1962

Question 8

Name the 4 different activation functions

Answer 8

Linear
Threshold Linear
Sigmoid
Binary Threshold

Question 9

Why cant a single layer network compute XOR.

Answer 9

It can only assign 1 weight to each input.

Question 10

Hidden layer

Answer 10

a layer of hidden units in an artificial neural network

Question 11

Why is the backpropagation algorithm not very biologically plausible?

Answer 11

There is no evidence error is propagated backwards in the brain, and nature never really provides feedback as detailed as the algorithm requires

Question 12

What aspect do multilayer networks have?

Answer 12

hidden units that are neither input units nor output units