L15 - Neural Networks 1: AI, Brain, Neural Computing

Question 1

What is the difference in computational capabilities between the brain and a computer?

Answer 1

Computer -> Extremely efficient performing serial tasks.

Brain -> Slower that computers, but far more parallel capability. Also deals well with noise.

Question 2

How many neurons does the human brain contain? How many other neurons is each connected to?

Answer 2

1. 100 billion
2. 10,000 connections via synapses

Question 3

How do neurons communicate?

Answer 3

Synapses

Question 4

What are the 3 components of a neuron?

Answer 4

Dendritic Tree -> Receives signals
Cell body -> Processes signals
Axon -> Transmits signals

Question 5

Describe the neural function…

Answer 5

Neuron: receives electric signal via dendrites

Cell body: processes the signal

Axon: transmits (outputs) signal, producing a pulse strength relative to the strength of the pulse received.

Question 6

Define the role of synapses and its functionality…

Answer 6

Chemical junctions that enable signal communication between Neurons.

Can excite or inhibit the dendrite (Neuron receiver).

Can release more neurotransmitter to enhance coupling between the 2 Neurons.

Question 7

What happens if a synapse excites or inhibits the receiving dendrite?

Answer 7

Excite -> Receiving Neuron is more likely to pass signal forward.

Inhibit -> Receiving Neuron is less likely to pass signal forward.

Question 8

What does synapse excitement vs inhibition determine?

Answer 8

The likelihood of the Neuron passing the signal forward.

Question 9

What is the motivation behind neural networks?

Answer 9

To replicate the human neural process onto a computer.

Question 10

What was the first development towards neural networks? Who developed it?

Answer 10

• Artificial Neuron -> Mimicked Neuron structure to process simple logical expressions.
• McCulloch and Pitts

Question 11

What are the components of the Artificial Neuron?

Answer 11

1. Inputs
2. Threshold Function
3. Output

Question 12

What is the Hebbian Learning theory?

Answer 12

1. When 2 Neurons fire together, their connection is strengthened.
2. The activity of Neuron firing is fundamental for learning and memory.
3. By strengthening Neuron connections, they will consistently fire together when given the same input.

Question 13

What is a Perceptron? Who theorised it?

Answer 13

1. A type of artificial Neuron that learns using weighted input connections. Each input is operated on with the associated weight before being input into threshold function.
2. Rosenblatt

Question 14

What is the main issue with the Perceptron? Why is this?

Answer 14

1. Can’t solve non-linear problems.
2. Can’t solve the XOR problem -> Only returns true if 01 or 10, but not if 11 (which we know is true).
3. This is because the classes in XOR aren’t linearly separable. We can’t draw a straight line to separate (0,0),(1,1) and (0,1),(1,0).

Question 15

What was the solution to the issue of perceptron non-linearity?

Answer 15

Multi-layer perceptron -> Implement hidden layers of Neurons between the input Neurons and the output.