IB catalogue 2026 FINAL - Flipbook - Page 15
A4.1
Machine learning fundamentals
Streamline your lesson planning;
the unit and chapter titles match
syllabus sections precisely to
save you time and enhance
learning efficiency. The resource
also provides flexibility in choice
of programming language to
cater to diverse teaching and
learning preferences.
What principals and approaches should be considered to ensure
machine learning models produce accurate results ethically?
SYLLABUS CONTENT
◆ Generative AI:
a form of artificial
intelligence capable of
generating text, images,
audio, video and other
digital artefacts, usually
in response to a prompt.
It is a form experiencing
rapid advances at the
time of writing.
◆ Machine learning:
a branch of AI where
computers learn from
data and experiences to
perform specific tasks or
solve specific problems,
without being explicitly
programmed to do so.
◆ Artificial
intelligence: computer
technology able to
perform tasks and
make decisions in a
manner that imitates
human intelligence.
There are two main
forms of AI: narrow (or
weak) AI is designed to
perform specific tasks
or solve specific types
of problems; general
(or strong) AI processes
human-level intelligence
and can operate across a
range of domains. While
speculation persists that
general AI is “close”, at
this time only narrow AI
technology is available.
206
By the end of this chapter, you should be able to:
A4.1.1 Describe the types of machine learning and their applications in the real world
A4.1.2 Describe the hardware requirements for various scenarios where machine learning
is deployed
A4.1.1 Types of machine learning
and their applications
TOK
What counts as knowledge?
Machine learning models “learn” from data, which raises questions about what constitutes
knowledge.
There are many opportunities to
make connections across the IB
Diploma with TOK link boxes and
ATL activities throughout.
Views on knowledge often distinguish between knowledge gained through experience (empirical)
and knowledge gained through reasoning (rational). Machine learning models acquire knowledge
empirically by processing vast amounts of data. However, unlike humans, machines do not
“understand” or reason about this data in the human sense. This raises the question: Can the
patterns and predictions that machines generate be considered “knowledge”, or are they simply
data-processed outputs?
Welcome to the world of machine learning! We live in a time of exciting growth and rapid
innovation in machine learning. Generative AI is making global headlines and has changed
the way we live and work in a very short timeframe. Speculation is rife that “general AI” is not
far from becoming reality. Certainly, it is an exciting topic, but what are machine learning and
artificial intelligence, and how do they work? Gaining an understanding of what is happening
behind the scenes is the goal of this chapter.
This chapter will not seek to dissect the details of the latest, greatest, news-making
developments in the field. That would be a fool’s errand as it would be obsolete before the
book is printed. Instead, the aim is to give you a solid understanding of the core theories and
techniques that form the basis of the entire field of machine learning. From these foundations,
you will be in a much stronger position to understand the true implications of modern
developments occurring in the field.
tip!
Before proceeding any further, it is important to clarify and differentiate Top
between
the terms
machine learning (ML) and artificial intelligence (AI). Artificial intelligence
is afollowing
broad field
Adapt the
as athat
guide to help determine which is the
best human
device for
a given scenario.
seeks to create systems capable of performing tasks that typically require
intelligence.
n For large and complete models, does it require real-time
processing?
n Yes: Consider GPUs for their parallel-processing
capabilities
A4
Machine learning
n No: TPUs might be a better choice for batch
processing with high efficiency in tensor operations
n For real-time inference (using a model for decision-making
after training), is the model deployed on edge devices?
n Yes: NPUs or ASICs, for optimized power and
efficiency
n No: Consider FPGAs for flexibility or ASICs for
efficiency if the task won’t change
Support students’ success
with essential tools,
including clear definitions
of key terms, practical
‘top tips’, cross-course
questions, and highlights of
common mistakes to avoid.
For models requiring future flexibility, are future
updates expected?
n Yes: FPGAs, due to their reprogrammability
n No: ASICs or GPUs, depending on whether the task is
more about speed or parallel processing
n Is low cost more important than cutting-edge
performance?
n Yes: Consider older generation GPUs or cloudbased solutions where hardware costs can be easily
absorbed
n Will there be a need to quickly scale processing power?
n Yes: Cloud GPUs or TPUs can offer scalable resources
as required
n
REVIEW QUESTIONS
1
A hospital is integrating a system that can automatically diagnose diseases from patientimaging data.
a
Describe whether this system should be classified as artificial intelligence, machine
learning or deep learning.
b Distinguish between regression-based and classification-based machine learning.
2
An email client uses a program to sort incoming emails into “Primary”, “Social”,
“Promotions” and “Spam” folders.
a
Identify whether this is an example of supervised or unsupervised learning.
b Describe your reasoning for this choice.
3
An autonomous vehicle company transfers the knowledge from a model trained in one city to
a new model designed to navigate another city.
a
Define “transfer learning”.
b Outline how this is an example of transfer learning.
c
Outline one possible limitation to the effectiveness of this approach.
d The original model was trained from thousands of hours of driving on roads under human
supervision to monitor and correct it when required. Describe the form of machine
learning used for the original model.
Build confidence through
engaging practical
activities, chapter
summaries, and targeted
review questions that
are designed to create a
deep understanding of
the subject matter.
4
A tech start-up is planning to deploy a large-scale machine learning system to predict stock
prices in real time.
a
Identify one type of hardware that would be critical for processing large volumes of realtime data in this context.
b Outline one reason that this type of hardware is suitable for real-time data processing in
machine learning applications.
c
5
Discuss one potential limitation of the identified hardware when used for
machine learning.
A university plans to implement an AI-driven system to analyse video lectures for enhancing
online learning experiences.
a
Identify two types of hardware that could be used for conducting machine learning
processing of video data in real time.
b For the two types of hardware identified, outline one possible reason for selecting each
device over the other.
214
A4 Machine learning
12
