What is a convolutional neural network (CNN)?
Length:
3 min
Published:
June 9, 2026
What is a convolutional neural network?
A convolutional neural network (CNN) is a type of neural network designed to work with images. Instead of looking at every pixel in isolation, it slides small filters across the picture to detect local patterns: an edge here, a corner there. Early layers find simple shapes, and later layers combine those into bigger features like an eye, a wheel, or a letter. The word convolution is just the name for that sliding-filter step.
This design is why CNNs dominate vision tasks. They assume that what matters in an image is local and can appear anywhere, so a filter that spots an edge in the top-left works just as well in the bottom-right. That makes them efficient and good at recognising the same object wherever it sits in the frame.
In plain words
Imagine looking at a photo through a small cardboard window, moving it across the picture bit by bit. Through each opening you notice tiny things: a line, a curve, a patch of colour. On its own each glimpse means little, but stacked together they tell you "this is a cat." A CNN does exactly this, with thousands of little windows running in parallel.
Where you see it
- Image recognition. Telling cats from dogs, reading handwriting, or sorting product photos.
- Medical imaging. Spotting tumours or fractures in scans, often as a second pair of eyes for doctors.
- Self-driving and cameras. Detecting lanes, pedestrians, and signs in a live video feed.
- Quality control. Catching defects on a production line faster than a human inspector.
Common pitfalls
- They need a lot of labelled data. A CNN learns from examples. Too few, and it memorises instead of generalising.
- Newer models are catching up. Vision transformers now match or beat CNNs on some tasks, so a CNN is not automatically the best choice.
- They are fooled by small tricks. Tiny, invisible changes to an image can flip a confident answer, which matters when safety is on the line.
- A right answer is not understanding. A CNN matches patterns; it has no concept of what a cat is. Treat its output as a strong guess, not a fact.
Related articles:
- What is a neural network? - The broader family that a CNN belongs to.
- Machine learning vs deep learning - Where CNNs sit and what makes a network "deep".
- What is multimodal AI? - Models that combine vision with text and other inputs.
Want to stay one step ahead?
Don't miss our best insights. No spam, just practical analyses, invitations to exclusive events, and podcast summaries delivered straight to your inbox.