# Computer Vision: Models, Learning, and Inference (2012)

### Part VII

### Appendices

**Appendix A**

**Notation**

This is a brief guide to the notational conventions used in this text.

**Scalars, vectors, and matrices**

We denote scalars by either small or capital letters *a, A, α*. We denote column vectors by bold small letters **a**, . When we need a row vector we usually present this as the transpose of a column vector **a*** ^{T}*,

*.*

^{T}We represent matrices by bold capital letters **B**, **Φ**. The *i** ^{th}* row and

*j*

*column of matrix*

^{th}**A**is written as

*a*

*. The*

_{ij}*j*

*column of matrix*

^{th}**A**is written as

*a**. When we need to refer to the*

_{j}*i*

*row of a matrix, we write this as*

^{th}*a*

*where the bullet • indicates that we are considering all possible values of the column index.*

_{i•}We concatenate two *D* × 1 column vectors horizontally as **a** = [**b**, **c**] to form the *D* × 2 matrix **A**. We concatenate two *D* × 1 column vectors vertically as **a** = [**b*** ^{T}*,

**c**

*]*

^{T}*to form the 2*

^{T}*D*× 1 vector

**a**. Although this notation is cumbersome, it allows us to represent vertical concatenations within a single line of text.

**Variables and parameters**

We denote variables with Roman letters **a**, **b**. The most common examples are the observed data which is always denoted by **x** and the state of the world which is always denoted by **w**. However, other hidden or latent variables are also represented by Roman letters. We denote parameters of the model by Greek letters ** μ**,

**Φ**,

*σ*

^{2}. These are distinguished from variables in that there is usually a single set of parameters that explains the relation between many sets of variables.

**Functions**

We write functions as a name, followed by square brackets that contain the arguments of the function. For example, log[*x*] returns the logarithm of the scalar variable *x*. Sometimes we will write a function with bullets • as arguments (e.g., atan2[•,•]) to focus the interest on the function itself rather than the arguments.

When the function returns one or more vector or matrix arguments, it is written in bold. For example, the function **aff**[**x**, **Φ**, ** τ**] applies an affine transformation to the 2D point

**x**with parameters

**Φ**,

*τ*and returns a new 2D vector output. When a function returns multiple outputs, we write this in Matlab notation so [

**U**,

**L**,

**V**] =

**svd**[

**X**] returns the three parts

**U**,

**L**,

**V**of the singular value decomposition of

**X**.

Some functions are used repeatedly throughout the text. These include:

• min_{x}*f*[*x*], which returns the minimum possible value of the function *f*[*x*] as we vary *x* over its entire valid range,

• argmin_{x}*f*[*x*], which returns the value of the argument *x* that minimizes *f*[*x*],

• max* _{x}* and argmax

*, which fulfill the same roles as min*

_{x}*and argmin*

_{x}*but where we are maximizing the function,*

_{x}• **diag**[**A**], which returns a column vector containing the elements on the diagonal of matrix **A**,

• *δ*[*x*] for continuous *x*, which is a Dirac delta function and has the key property *∫* *f*[*x*] *δ*[*x*−*x*_{0}]*dx* = *f*[*x*_{0}],

• *δ*[*x*] for discrete *x*, which returns 1 when the argument *x* is 0 and returns 0 otherwise, and

• heaviside[*x*], which represents the Heaviside step function. It returns 0 when the argument *x* < 0 and returns 1 otherwise.

**Probability distributions**

We write the probability of a random variable *x* as *Pr*(*x*). We write the joint probability of two variables *a,b* as *Pr*(*a, b*) and the conditional probability of *a* given *b* as *Pr*(*a*|*b*). Sometimes, we wish to specify the exact value *b** that *a* is conditioned upon and here we write *Pr*(*a*|*b* = *b**). Occasionally, we denote that variables *a* and *b* are independent by writing *a* ⊥⊥ *b*. Similarly we indicate that *a* and *b* are conditionally independent given *c* by writing *a* ⊥⊥ *b*|*c*.

Probability distributions are written in the style *Pr*(**x**|** μ**,

**∑**) = Norm

_{x}[

**,**

*μ***∑**]. This returns the value of the multivariate normal distribution for data

**x**when the distribution has mean

**and covariance**

*μ***∑**. In this way, we always distinguish the argument of the distribution (here

**x**) from the parameters (here

**,**

*μ***∑**).

**Sets**

We denote sets with calligraphic letters . The notation ⊂ indicates that is a subset of . The notation *x* ∈ indicates that *x* is a member of the set . The notation = ∪ indicates that set is the union of sets and . The notation = \ indicates that set consists of all of the elements of except those that are in .

Often we write out a set explicitly in terms of the elements and for this we use curly brackets so that = {*x,y,z*} indicates that the set contains *x,y*, and *z* and nothing else. When a set is empty, we write = {}. We use the notation {*x** _{i}*}

^{I}

_{i}_{=1}as shorthand to represent the set {

*x*

_{1},

*x*

_{2},…,

*x*

*}, and we may write the same set in the compact form*

_{I}*x*

_{1…I}if it is part of an equation.