iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A convolutional neural network (CNN), also called a ConvNet, is a feed-forward neural network with at least one convolutional layer. That layer applies learned filters across an input to detect patterns. CNNs are especially useful for structured, grid-like data such as images and regularly sampled time series.
How does a convolutional neural network work?
A convolutional layer uses a small set of learned values called a filter, kernel, or feature detector. The filter is applied at different positions across the input. Because the same filter is reused, the network can respond to a pattern wherever it appears instead of learning an entirely separate set of weights for every location.
During training, the network adjusts filter values so they respond to patterns useful for its task. In supervised image classification, for example, it can learn from labeled images and their categories using backpropagation. Later layers combine earlier responses into more complex representations.
In CNN software, the operation called convolution may not be strict mathematical convolution: depending on the convention, it may instead be cross-correlation, which does not flip the filter. The name remains common in machine learning. The Deep Learning textbook explains this distinction.
#1 Best Overall
What is a convolutional layer?
A convolutional layer is the part of a neural network that applies one or more learned filters across an input. Each filter detects a pattern in local portions of that input, and its responses form feature maps that later layers can use. The defining feature of a CNN is the presence of at least one such layer; pooling is common, but it is not required by the definition. NIST’s AI 100-2e2025 glossary defines CNNs as feed-forward neural networks that include at least one convolutional layer.
What are CNNs used for?
CNNs are widely used for visual recognition, including tasks that ask what an image contains, where objects are, or which pixels belong to a particular region. These are commonly called image classification, object detection, and image segmentation. CNNs can also process other structured signals, including one-dimensional time series and applications involving speech or handwriting.
Rank #2
The architecture alone does not guarantee accuracy, interpretability, or robustness. Whether a particular CNN is suitable depends on the task, input data, and model.
Recommended Free Tools
What makes CNNs different from networks without convolution?
The key choices are local processing and shared weights: a filter looks at a local portion of the input, then reuses the same learned values at other positions. This fits data where nearby elements have meaningful relationships and where useful patterns may appear in different locations. Images are a familiar example because pixels occupy a two-dimensional grid.
Rank #3
Pooling is often included in CNN designs, but it is an architectural choice rather than a requirement. CNNs also do not have one fixed sequence of layers; their design depends on the problem.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Further reading
For a deeper technical explanation, see the convolutional networks chapter of the Deep Learning textbook by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. For a visual-recognition course perspective, Stanford’s CS231n materials cover CNNs and practical training.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
