A Convolutional Neural Network (CNN) is a class of deep neural network specialized for processing grid-like data such as images. CNNs are composed of layers including convolutional layers (which apply learned filters/kernels that respond to local patterns like edges), activation layers (introducing non-linearity, e.g. ReLU), pooling layers (down-sampling spatial resolution to achieve translational invariance and reduce computation), and fully connected layers (to aggregate features into final predictions). This architecture exploits spatial locality by learning filters that act on small patches and are shared across the image, making CNNs very effective for image classification, object detection, and other computer vision tasks. Notable CNN architectures include LeNet, AlexNet, VGG, ResNet, etc., each introducing innovations in depth and connectivity to improve performance.