It earns a place when
- Use CNNs when local spatial structure and translation patterns matter, especially in images and some signals.
- You can evaluate it against a credible baseline.
- Its output fits the decision and data constraints.
Machine learning models · Phase 4 · Lesson 32
Convolutional neural networks
Understand convolution, receptive fields, weight sharing and pooling in image models.
Lesson goal
Understand convolution, receptive fields, weight sharing and pooling in image models.
The core idea
A convolutional filter scans local patches and reuses the same detector across the image. Early filters can detect edges; later layers combine them into shapes.
Mental model
Picture it this way. A convolutional filter scans local patches and reuses the same detector across the image. Early filters can detect edges; later layers combine them into shapes. The important question is what assumption this picture makes, and whether that assumption fits the data.
Mathematical core
A convolution computes local weighted sums with shared parameters. Stride changes spatial resolution; padding controls borders; pooling or strided convolution reduces resolution.
Worked example
A classifier for Australian road signs can detect local edge patterns in one location and reuse the detector wherever the sign appears.
When to use it
Do not make it the default when
- Do not flatten an image too early if spatial locality is important, and do not assume convolution alone handles all global relationships.
- A simpler model has not been tested.
- The data or target definition is still unclear.
Failure modes
Watch for this. Data leakage through near-duplicate images, class imbalance and distribution shift across cameras can dominate architecture choice.
When a result looks surprisingly good, inspect the split, target timing, error slices and data-generating process before celebrating.
Practice
Use this as a small experiment rather than a recipe to copy blindly. Change one thing, record the result and explain the change.
Do this. Use a small CNN on a standard image dataset. Visualise a few learned feature maps and compare it with a dense network of similar parameter count.
Retrieval check
Answer from memory first. The buttons reveal feedback, but the durable step is explaining why.
1. What does weight sharing mean in a convolution?
2. What does a receptive field describe?
3. Why can near-duplicate images leak?
Transfer prompt. Describe one real problem where this model or idea would be a sensible candidate. Name the target, the main risk and the metric you would inspect.
Primary source
CS231n. Use the source for the deeper treatment after you can explain the lesson's core idea without looking.