AeroLab Reading room

CS / A2 level

Artificial intelligence and learning

AI includes search, knowledge representation and learning methods. Heuristic search uses a guide to explore possibilities; a heuristic’s properties affect optimality and efficiency.

Learning objectives

The big idea

AI includes search, knowledge representation and learning methods. Heuristic search uses a guide to explore possibilities; a heuristic’s properties affect optimality and efficiency.

Supervised learning uses labelled examples, while unsupervised methods seek patterns without those labels. Separate training from evaluation, consider overfitting and assess errors across relevant groups rather than reporting accuracy alone.

Worked example

A classifier gets 90 of 100 test cases correct. Find accuracy.

Follow the solution

Accuracy=90/100×100=90%; this alone does not reveal class-specific errors.

Investigate and explain

Predict the worked result before calculating. Change one input, record the new result, and explain which assumptions still hold.

Use the prediction activity below; a local model appears only where it supports this topic.

Open the interactive investigation

Check your understanding

1. What is overfitting?

  1. Always learning too slowly
  2. Using too little memory
  3. Learning training-specific patterns that generalise poorly
Check the answer and reasoning

Learning training-specific patterns that generalise poorly

A model can fit training examples well yet fail on new data.

2. A classifier gets 90 of 100 test cases correct. Find accuracy.

  1. 90%
  2. 10%
  3. 0.9%
Check the answer and reasoning

90%

Accuracy=90/100×100=90%; this alone does not reveal class-specific errors.

3. May test data be repeatedly used to tune a model without affecting evaluation?

  1. Yes
  2. No
  3. Only for small data
Check the answer and reasoning

No

Repeated tuning leaks information from the test set, weakening its value as independent evidence.