Lesson 39 min

CNNs

Convolutional Neural Networks — the workhorse of image AI.

What you will learn

✓How CNNs stack convolution layers
✓What pooling does
✓Why early vs late layers differ

Explanation

A CNN stacks many convolution layers, each learning its own filters. Pooling layers shrink the image between steps, keeping the important signal and cutting computation.

A beautiful pattern emerges: early layers learn simple features (edges, colours), middle layers learn parts (eyes, wheels), and final layers recognise whole objects.

This hierarchy is why CNNs are so effective at image classification.

Real-world use

AlexNet, the 2012 model that kicked off the deep-learning boom, was a CNN.

Common mistakes

• Assuming every layer detects whole objects — early layers only see simple edges.

Practice

Explain the early-to-late layer hierarchy of a CNN in your own words.

Knowledge check

0/2 answered

1. In a CNN, early layers typically detect...

2. Pooling layers mainly...

Answer all questions to check.

Previous Next