Intro / Neural Systems

Question 1

Total venture capital money for pure AI startups 2014

Answer 1

$300million

Question 2

Venture capital investment in robots in 2015

Answer 2

$587million (double the amount of 2014)

Question 3

Percent of people in UK who have heard the term ‘machine learning’? Where from?

Answer 3

9%

75% mainstream media, 21% entertainment

Question 4

Simple low tech example of Bio-inspired tech from burrs?

Answer 4

Velcro

Question 5

How many neurons in your brain?

Answer 5

86 billion

Question 6

How many synapses in your brain

Answer 6

10 trillion

Question 7

How many possible calculations per second?

Answer 7

10 quadrillion

Question 8

Name two Brain projects

Answer 8

BRAIN Initiative $300+ million

Human Brain Project €1 billion over 10 yrs

Question 9

What increase needed to bridge gap between best computers and human brain?

Answer 9

100,000 times

Question 10

Name some parts of a real neuron

Answer 10

Dendrites, Axon, Soma

Question 11

Name parts of Synapse

Answer 11

Axon, Synaptic cleft, neurotransmitters, receptor molecules, dendrite

Question 12

Two main types of interneurons

Answer 12

Inhibitory, excitatory

Question 13

Three examples of output functions

Answer 13

Identity, step, sigmoid

Question 14

Hebb’s rule is used when

Answer 14

updating weights in learning

Question 15

Hebb’s rule suffers from

Answer 15

self-amplification (unbounded growth of weights)

Question 16

A neural network with a single layer is called a

Answer 16

perceptron

Question 17

Can a single perceptron separate XOR inputs?

Answer 17

No, not linearly separable, need multilayer

Question 18

The three types of units in a multilayer neural networks?

Answer 18

Input, hidden and output

Question 19

Which output function sound should a multi-layer network no use?

Answer 19

Multi-layer networks should not use linear output functions because a linear transformation remains a linear transformation. Therefore, such a network would be equivalent to a networks with a single layer.

Question 20

In an MLP how is the error of the hidden units found?

Answer 20

Backpropagation of error

Question 21

Eight steps in backpropagation of error

Answer 21

1. Initialize weights (random, around 0)
2. Present pattern
3. Compute hidden
4. Compute output
5. Compute delta output
6. Compute delta hidden
7. Compute weight changes
8. Update weights (back to 2)

Question 22

Name a solution to overfitting

Answer 22

Use a validation set