What is convolution in deep learning?

Convolution is an operation that applies a small learnable filter across an input to detect local patterns.

In image-based deep learning, the filter moves over small regions of the image and computes a weighted combination of nearby pixel values.

Input image
    ↓
Small filter slides across it
    ↓
Local responses are computed
    ↓
Feature map

Simple example

Suppose a 3 × 3 filter is applied to a small 3 × 3 region of an image.

The filter multiplies its values with the corresponding image values and adds the results together:

output = Σ(input_value × filter_weight) + bias

That produces one value in the output feature map. The filter then moves to the next position and repeats the same calculation.

Why is convolution useful?

A convolutional layer has two important properties:

• Local connectivity: each output depends only on a small nearby region of the input.
• Weight sharing: the same filter is reused across different positions.

This allows a CNN to detect the same pattern wherever it appears in an image while using far fewer parameters than a fully connected layer.

Different filters learn different patterns, producing multiple feature maps.

For the complete CNN flow, see “How does a CNN work?”. For the architecture itself, see “What is a convolutional neural network (CNN)?”.

In short, convolution scans an input with shared learnable filters to detect local patterns and produce feature maps.