Quiz

Gaussian Processes provide the theoretical foundation for this course. This quiz gives you the opportunity to check if you have understood the main properties of Gaussian Processes that we need in this course.

Question 1

Why do we prefer to use the concept of a Gaussian Process rather than a multivariate normal distribution, even in the case of modelling shapes that are defined using only a discrete representation?

Because the Gaussian Process makes it clear that we are working with a distribution over functions (or vector-fields).

Because a Gaussian Process is more flexible than a multivariate normal distribution.

Because a multivariate normal distribution cannot be used to represent functions.

Question 2

Why do we define a Gaussian Process model over deformations and not shapes?

To make the model independent of rotation and translation.

The distribution of points cannot be modelled directly.

We are interested in modelling the differences between shapes, which are naturally represented as deformation fields.

Question 3

Assume that you are modelling a 3D shape using a shape representation of 15 points. What would be the dimensionality of the covariance matrix of the joint normal distribution?

15 x 15

30 x 30

45 x 45

Question 4

How does the covariance function for a vector-valued Gaussian Process differ from a scalar-valued Gaussian Process?

The covariance function k(x,x’) is scalar-valued.

The covariance function k(x,x’) is vector-valued.

The covariance function k(x,x’) is matrix-valued.

Question 5

What can we say about the deformations if the covariance function k(x,x’) is non-zero only if x = x’?

The deformations of any two distinct points are independent.

The deformations of any two distinct points are correlated.

Only some of the points are correlated.