Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
28 Cards in this Set
- Front
- Back
When is classical AI suitable? |
Well-defined tasks for which there are a clear set of rules |
|
When is classical AI not suitable? |
Situational awareness and comparisons based on previously learnt experiences |
|
What is modern AI based on? |
Building blocks of the brain |
|
How does a neuron receive input? |
A neuron receives inputs from other neurons via its dendrites |
|
What does a neuron generate with sufficient stimulation? |
An impulse / action potential |
|
How do impulses travel? |
Impulses travel across the axon of a neuron which is connected to other neurons via synapses ("junctions" in the brain) |
|
What are the four main components of a MCP (McCulloch-Pitts) modelled neuron? |
Weights, bias, threshold, output |
|
How are weights interpreted in the MCP model? |
Positive weights are considered excitatory, negative weights are considered inhibitory. |
|
What is the role of the bias in the MCP model? |
The bias sets the firing threshold for the neuron
|
|
How many inputs does an MCP neuron have? |
Two (x, y) |
|
How does an MCP neuron decide to fire? |
Input (x, y) is multiplied by their respective weights and added together. If total >= threshold then the neuron fires. |
|
What is the formula for calculating whether or not an MCP neuron fires? |
xW1 + yW2 + b >= 0 |
|
What is the purpose of a perceptron? |
Classify data into different classes. Output of 1 if input falls into a class, output of 0 if falls into another class |
|
How to plot the feature-space line? |
xW1 + yW2 + b = 0 can be rearranged in the form y = mx + c |
|
How does a perceptron calculate correct bias and weight values? |
Perceptron is given training data containing expected outputs for some inputs. Bias and weight values are then adjusted accordingly until training data is accepted |
|
What formula is used to adjust the weights of a perceptron? |
Delta Wi = n(Tp - Op)Ii |
|
What formula is used to adjust the bias of a perceptron? |
Delta b = n(Tp - Op) |
|
What are the limits of a MCP neuron? |
An MCP neuron can only separate between two classes
|
|
How are the limits of a MCP neuron dealt with? |
Neurons are connected together to form a neural network |
|
What does a binary discriminant neuron (BDN) do? |
Classifies input data into one of two responses (true or false, 1 or 0) |
|
How many input patterns can a BDN classify? |
2^n patterns, where n is the number of inputs |
|
What is the number of inputs to address the RAM neuron known as? |
A tuple. Eg: if there are 8 bits, it would be an 8 tuple |
|
Why isn't a unary RAM chip used? |
Not cost efficient, can't discriminate between learning sets and actual data |
|
What is the solution to the problems of using a unary RAM chip? |
Arrays of n-tuple RAM chips are used. Large input vectors are divided between these chips |
|
What is a compound RAM also known as? |
A class discriminator |
|
How many class discriminators would we need to learn to recognise digits 0-9? |
10 class discriminators (one for each digit) |
|
What are the memory requirements of a class discriminator? |
In a discriminator with k times over-sampling of an input vector of size R and a tuple size of n, M RAM's are needed, where M = (k*R)/n Memory size of each RAM must be 2^n bits Total memory required is M*(2^n) |
|
What does a small tuple size imply? |
Each neuron behaviour is decsribed by a small list of mappings |