praxis.ibLog in
A4.3 Machine learning approachesA4.3.9

How CNNs are designed to adaptively learn spatial hierarchies of features in images

Describe how CNNs are designed to adaptively learn spatial hierarchies of features in images

AHL≈ 55 minCommand term: DescribePremium

In this lesson

  • Describing the basic architecture of a CNN: input, convolutional, activation, pooling, fully connected and output layers
  • Describing how convolution and pooling extract features, and how stacked layers learn a hierarchy of features
  • Describing the effect of the number of layers, kernel size and stride, activation function and loss function

This lesson is part of Premium

Every topic has one free lesson. Premium unlocks every interactive lesson, the exam-style questions at the end of each one, and the rest of Praxis IB.

Log in or sign upTry the free lesson: A4.3.1 Linear regression for predicting continuous outcomes

What's inside

  1. LearnWhy images need a different network
  2. LearnThe architecture of a CNN
  3. LabConvolve, activate, pool
  4. CheckQuick check
  5. LearnLearning a hierarchy of features
  6. CheckOrder the steps
  7. LearnDesign choices and their effects
  8. CheckMatch each change to its main effect
  9. ExamWorked exam answer
  10. ReviewMistakes examiners see
  11. Exam practiceExam-style questions

Ends with 4 exam-style questions worth 11 marks, with a self-marking markscheme.