Comp304 Practice Quiz

A vibrant illustration depicting self-organizing maps, neural networks, and data clustering, with elements of circuitry and mathematical symbols in a modern, digital style.

Comp304 Practice Quiz

Welcome to the Comp304 Practice Quiz! Test your knowledge on the fascinating world of Self-Organizing Maps (SOMs) and their applications in neural networks. This quiz is designed for students and enthusiasts alike, helping you to solidify your understanding and explore various concepts.

Key Features:

12 challenging questions
Covering a range of topics from SOM training to node classification
Instant feedback on your answers

12 Questions3 MinutesCreated by LearningMap101

Which of the following is true regarding training a SOM

During training the SOM will classify data according to similar properties through adjusting weights linearly using a gaussian function

SOM can use a "Supervised" learning approach applying backpropagation to adjust weights of every node in the network

In vector quantization the training set is supervised through a method called learning vector quantization(LVQ) for fine tuning the SOM

SOM performs dimensionality reduction, cluster labeling as well as reorganize instances

SOM can not be trained since it is not e neural network

None of the above

A supervised version of SOM(if it exists) require that vectors used for training have an unknown classification

True

False

Can not say

Which of the following is true regarding selecting the BMU

The input neuron with dimensions closest to the origin(0,0) when reduced from 3D is the BMU

The input neuron with minimal distance to all of the given nodes is the BMU

The value of the gaussian function for the BMU is always 1

An input neuron with the highest weights is the BMU

The input neuron with the lowest weights is the BMU

The output neuron/node mapped to the 1st input with the smallest weights will always be the BMU

Which of the following is not false regarding a SOM neural network

If node A and node B have 150 neurons between them, node A would still be able to offer feedback to Node B as long as they are within the same neural grid/lattice

Feedforward networks and SOM treat the concept of neighbourhood the same way

Using random weights for SOM outputs is not applicable, a more structured approach of calculating weights must always be implemented from the very begining

An infinite number of iterations is acceptable for training an SOM with a large number of input and output neurons

None of the above

Which of the following is true regarding the degree of training each output node gets

The BMU does not receive any training since it is the standard every other node is trying to match up to

The nodes furthest to the BMU receive most training since they are not as effective in mapping inputs

The gaussian function penalizes nodes that are 10 nodes away from the BMU

A Mexican hat function can be used to calculate the degree of training each node gets

None of the above

Can not say

The neighbours of the BMU always learn more from the training pattern since the value of their gaussian function is always higher

Which of the following is not false regarding the learning rate of an SOM

Increases monotonically with each iteration

Will always be 0.5 for maximum effectiveness on forming clusters of the SOM

When training an SOM it is always important to make sure a learning rate is always reduced to 0.02 for nodes with similar weights and 0.5 for the rest

The aim of training the SOM, is to make the node weights exactly the same as the training elements/ inputs through using the highest possible learning rate

When the learning rate is approximately zero, it is ok to stop the training process/

Can not say

When using the Learning Vector quantisation, weights for the winner are calculated the same way for correctly classified inputs as well as misclassified inputs

True

False

Can not say

Comp304 Practice Quiz

Comp304 Practice Quiz

More Quizzes