Examhaus — Achieve with confidence.

IB Computer Science A4.1: Machine learning fundamentals

New IB Computer Science syllabus · first exams May 2027 · Theme A · SL & HL

Guiding question: What principles and approaches should be considered to ensure machine learning models produce accurate results ethically?

A4 Machine learning takes 5 hours at SL and 18 hours at HL. SL students study A4.1 and A4.4; HL students also study A4.2 and A4.3. A4.1 covers the main types of machine learning and the hardware they run on.

StatementTopicCommand term
A4.1.1Types of machine learning and their applicationsDescribe
A4.1.2Hardware requirements for ML scenariosDescribe

Key points

A4.1.1 Types of machine learning

  • Supervised learning: trains on labelled data (with known correct outputs) to classify or predict. Example: medical imaging diagnosis.
  • Unsupervised learning: finds structure in unlabelled data, such as clusters or associations. Example: market basket analysis.
  • Reinforcement learning: an agent acts in an environment and receives rewards or penalties, learning a policy by trial and error. Example: robot navigation.
  • Deep learning: neural networks with many layers. It's a technique used across supervised, unsupervised and reinforcement learning. Examples: object detection, natural language processing.
  • Transfer learning: takes a model already trained on a large general dataset and fine-tunes it for a new, specific task. It needs less data and less training time.

Applications named in the guide: market basket analysis, medical imaging diagnostics, natural language processing, object detection and classification, robotics navigation, and sentiment analysis.

A4.1.2 Hardware for machine learning

Choose hardware based on processing, storage and scalability:

  • Standard laptops: fine for learning and small models.
  • GPUs: massive parallelism for training.
  • TPUs: chips designed specifically for neural-network maths.
  • FPGAs: chips that can be reprogrammed after manufacture.
  • ASICs: chips fixed for one purpose. Very efficient, as in phone face unlock.
  • Edge devices: run models locally for low latency and privacy.
  • Cloud platforms: rent scalable compute on demand.
  • HPC centres: supercomputing for the largest models.

Exam-style question

Describe the difference between supervised and unsupervised learning, with one real-world example of each. [6]

Mark scheme

  • Supervised [3]:
    • trained on data with known correct labels [1]
    • learns the mapping from input to output by minimising prediction error [1]
    • example: classifying spam email, or medical imaging diagnosis [1]
  • Unsupervised [3]:
    • trained on data with no labels [1]
    • discovers internal structure such as clusters or associations [1]
    • example: customer segmentation, or market basket analysis [1]

Common mistakes that cost marks

  • Treating AI and ML as the same thing. AI is the broad field, ML is part of AI, and deep learning is part of ML.
  • Equating deep learning with supervised learning. Deep learning is a technique, used in all three learning types.
  • Saying unsupervised learning needs no data. It needs plenty of data, just no labels.
  • Choosing hardware without justifying it against processing, storage and scalability for the specific scenario.

Frequently asked questions

Which parts of A4 do SL students study?
A4.1 (machine learning fundamentals) and A4.4 (ethical considerations). A4.2 and A4.3 are HL only.

What's the difference between an FPGA and an ASIC?
An FPGA can be reprogrammed after it's made; an ASIC is permanently built for one task, which makes it faster and more efficient but inflexible.

Go further