Deep learning / artificial neural network · BEGINNER LESSON
Convolutional neural network
A convolutional neural network (CNN) is a type of artificial neural network designed to process grid-like data such as images by learning features with convolution, pooling, and fully connected layers. It measures patterns in the input through learned filters or kernels, then uses those features for tasks such as classification, object detection, or segmentation.
What it is and what it measures
A convolutional neural network (CNN) is a deep learning model and a subset of artificial neural networks (ANNs). It is built to process structured, grid-like data such as images. Instead of measuring one handcrafted rule at a time, it learns filters, also called kernels, that respond to local patterns in the input, such as edges, textures, shapes, or object parts. The result is a set of feature maps, which are arrays that show where each learned pattern appears in the input.
[1] Convolutional neural networks (CNNs): concepts and applications ... [2] 7.4 Convolutional Neural Networks - Principles of Data ... [3] An Extensive Study of Convolutional Neural Networks - PMC - NIH
How it works
A typical CNN has convolutional layers, pooling layers, and fully connected layers. In a convolutional layer, a kernel moves across the input and performs element-wise multiplication to produce an activation or feature map. An activation function such as rectified linear unit (ReLU), a nonlinear function that keeps the model from being only linear, may follow. Pooling layers down-sample the feature maps, which reduces spatial size and parameters while keeping important information. Fully connected layers at the end combine the learned features to make a prediction, such as a class label. CNNs often use weight sharing and local connectivity, which help reduce the number of parameters compared with fully connected networks that connect everything to everything.
[1] Convolutional neural networks (CNNs): concepts and applications ... [2] 7.4 Convolutional Neural Networks - Principles of Data ... [3] An Extensive Study of Convolutional Neural Networks - PMC - NIH
Training and learning
Training means finding good values for the kernels, weights, and biases. A forward pass sends input through the network to produce an output. A loss function then measures error by comparing the output with the target. Backpropagation uses gradient descent, a method that changes parameters to reduce loss, and the process repeats until performance improves. Lower layers tend to learn simple features first, while deeper layers learn more abstract features.
[1] Convolutional neural networks (CNNs): concepts and applications ... [3] An Extensive Study of Convolutional Neural Networks - PMC - NIH
Limitations
CNNs are powerful, but they are not simple to use well. They can overfit when the model has too many parameters relative to the data. They often need large datasets, substantial computing power, and careful choice of hyperparameters such as number of layers, learning rate, and kernel size. Their internal reasoning can also be hard to interpret, so they are sometimes described as a black box. Data inconsistencies and poor model tuning can reduce output quality.
[1] Convolutional neural networks (CNNs): concepts and applications ... [3] An Extensive Study of Convolutional Neural Networks - PMC - NIH
Why they are useful
CNNs are widely used in computer vision because they capture spatial relationships in images. Reported uses include image classification, object detection, locating objects, edge detection, and semantic segmentation, which assigns a class to each pixel.
[2] 7.4 Convolutional Neural Networks - Principles of Data ... [3] An Extensive Study of Convolutional Neural Networks - PMC - NIH
SEE THE IDEA
How a CNN turns an image into a prediction
Grid-like data made of pixel values.
Kernels scan local regions to detect patterns and make feature maps.
Down-samples feature maps to reduce size while keeping important signals.
Combine learned features to produce the final class or score.
The input is transformed step by step: local patterns are detected, compressed, and then combined for a final prediction.
A CNN learns features from local image regions and uses them to make a decision.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.