The theoretical minimum mathematics you

need to understand AI
While it might not seem like it, an understanding of mathematics is essential in having an
insight into the foundations of AI and how AI can be used to transform. In this document, I
will go over the basic mathematical concepts, how they work, and real-world applications
and practical implications. They are split into three parts: Linear Algebra, Calculus,
Probability and Statistics. Knowledge about basic algebra is needed to fully get to grasp
with this document.

Linear Algebra rotating, and projection on the

data.
This topic is the primary mathematical
concept used in Artificial Intelligence,
and has three objects with specific
properties:
• Scalars – This is a single number
with only a magnitude and no
direction. An example is mass or
energy.
• Vectors – This is a number with
both magnitude and direction.
You might consider them as
points in space with each number
corresponding to a coordinate on
an axis.
• Matrices – This is an array of
numbers that will contain x rows
and y columns.
• Eigenvalues and Eigenvectors –
This is the scalar and vector
quantities associated with the
linear transformation of matrices.

We can then use these components and

apply them to data and AI. For example:

• Vectors can be used to represent

data points or other quantities.
• Matrices store large amount of
data such as images concisely and
allow for manipulation.
• Linear transformations permit
operations, such as scaling,
 Calculus

Calculus describes changes in applied to vectors and matrices:

parameters, functions, errors, and Gradient, Jacobian, Hessian, and
approximations. The following are the Laplacian.
most important concepts that relate to • Gradient Algorithms – By using
AI. differentiation, we can use
optimisation algorithms.
• Derivative – This is the rate of
Optimisation algorithms refers to
change of one variable with
the task of minimising or
respect to another variable. An
maximising an objective function,
example of this would be a
and this is done by finding the
gradient, as it is the change in y by
global minimum/maximum.
the change in x. These derivatives
Gradient descent is the most
can be added, subtracted
common and is a first order
multiplied, and divided using
algorithm and therefore only
different rules. This then can also
considers the first derivative. On
be applied to work out the
each iteration, the parameters are
derivatives of other mathematical
updated based on where the
functions like sine, cosine, and
gradient gives the direction of the
logarithmic functions. Finding the
steepest ascent. We can then
derivative of something is known
• follow the direction of the slope
as differentiation.
until the minimum or maximum
• Integral – This is the reverse of
has been found.
differentiation and is used to find
areas and volumes. As this is the We can the use calculus in real-world
reverse of differentiation, the applications in AI:
operations are done in reverse.
• Differentiation allows us to find
Finding the integral is known and
the optimal weights for machine
integration.
learning models.
• Vector and Matrix Calculus – We
• Integration helps compute area
can find the derivatives of vectors
and volume which can be useful in
and matrices. There are different
AI.
derivative operators that can be
 Probability and Statistics

heads, then we can say that there

are four outcomes for this
Probability and statistics allow us to
variable: 0 Heads, 1 Head, 2
analyse and understand data, which then
Heads, and 3 Heads. However,
in turn, allows us to make informed
there are eight possible
decisions and predictions in AI.
combinations of coins that can
• Basic Statistics – This includes fall, and this way we are able to
stats such as mean, median, calculate the probability of the
mode, variance, standard number of heads that we may get.
deviation and so on. All of these • Common Distributions – This is
provide an average of sorts, but the distribution of the probability
they are calculated differently. of an event happening in
For example, mean is the sum of graphical form. The x-axis is the
all the numbers in the dataset event while the y-axis is the
divided by the number of probability density. There are
numbers in the dataset. Median is many types of distributions such
the number in the dataset that is as normal, gaussian, and
exactly in the middle between the exponential.
numbers, and mode is the
number in the dataset that
appears most often. All of them
have a different use and
application.
• Probability Rules – There are four
probability rules that
mathematics follow: Addition
Rule (where one or both of the
events will occur), Multiplication
Rule (where both of the events
This is used in AI in the following ways:
occur), Conditional Probability
(where one event happens given • Statistics allow us to summarise
that the other event has and describe the major features
happened) and Complementary of the dataset.
Rule (where one or the other • Inferential statistics allow us to
event occurs, and they are draw conclusions and make
mutually exclusive). predictions based on samples.
• Random Variables – A random • Probability is used for modelling
variable is a set of values from an uncertainty and randomness in AI.
experiment. For example, if we
flipped three coins at the same
time, and we assigned the
random variable as the number of