A short quiz on machine learning in computer vision

Adrian F. Clark

1. What is the most common training techinque for a neural network?

   • forward propagation

   • back propagation

   • back and forward propoagation
   • sideways propagation

   The most widely-used training technique is called back-propagation. It is a way of adjusting the weights of each neuron to reduce the size of the error when evaluating inputs.

2. What is genetic programming?

   • the execution of parse trees

   • the use of a genetic algorithm to 'write' a program

   • a program that processes genetic data
   • a program that implements a genetic algorithm

   Genetic programming is the use of a genetic algorithm to 'write' programs. It combines program fragments in the form of parse trees into these programs, which it evaluates by running them.

3. How does SVM work?

   • it reduces the problem to a linear one and finds the best straight line

   • it finds the hyperplane that maximizes the distance between classes

   • it finds the smallest hypersphere that minimises encloses each class
   • it finds the hyperplane that minimizes the distance between classes

   SVM maximizes the distance between classes and uses a hyperplane to separate them.

4. Which of the following is a deterministic machine learning technique?

   • SVM

   • WISARD
   • MLP
   • GP

   All the techniques listed except SVM are stochasic, meaning they involve the use of random numbers and hence normally produce different results every time they are run on the same data. SVM doesn't do this and so is deterministic, producing the same output every time the same input is used.

5. In Genetic Programming, programs are commonly presented in which language?

   • English
   • Python

   • Lisp

   • Java

   Programs in GP are normally presented as Lisp. In fact, in the author's GP system, programs are converted to Lisp and executed in that form: the overhead of doing this is tiny.

6. Why might one "polish" the result of a genetic algorithm?

   • because a numerical optimizer is too inaccurate
   • because the GA gives the wrong result

   • to improve the accuracy of the solution

   • because the GA gives its result in the wrong coordinate system

   A GA is excellent at exploring the breadth of a search space and finding an approximate solution but it is less effective at determining the exact solution. Using the GA's result to initialize a numerical optimizer, which is what "polishing" means here, combines the best of both algorithms.

7. What is the purpose of the RELU layer in a CNN?

   • it determines the final result
   • it down-samples regions of its input
   • it applies the convolution coefficients to an image

   • it applies the activation function

   The RELU layer applies the activation function to each element independently.

8. What is the purpose of the POOL layer in a CNN?

   • it determines the final result

   • it down-samples regions of its input

   • it applies the convolution coefficients to an image
   • it applies the activation function

   The POOL layer reduces the size of its inputs, usually by down-sampling but sometimes by averaging.

9. What type of images does WISARD need?

   • 8 bits/pixel images
   • floating point images

   • binary images containing only the values zero and unity

   • colour images

   WISARD uses a single bit from each image location to set or clear a bit in its register so it requires binary images. The diagram in figure 10.2 of the lecture notes makes this clear.

10. In a deep neural network, what does the 'deep' refer to?

    • it takes many inputs
    • it allows many generations when training

    • it has many hidden layers

    • it produces many outputs

    A 'deep' network is one with many hidden layers. Modern networks often have >100 hidden layers...and take a Very Long time to train.