A convolutional neural network (CNN, or ConvNet) is a class of deep, feed-forward artificial neural networks that has successfully been applied to analyzing visual imagery. It falls under the [deep-learning] tag.
Convolutional Neural Networks (CNN) are a type of neural-networks used in many computer-vision problems. This architecture allows sharing the set of weights between different spatial regions in the input. Training a CNN for solving a machine learning problem such as object recognition involves finding the decision boundaries for classification as well as learning the kernels of the filters that generate the features without prior manual tuning.
In a typical CNN the input is convolved with multiple filters to produce feature maps. A pooling layer selects which responses to route further into the network. Non-linear transformations are applied to responses similar to non-linearities found in multi-layer perceptron.
The idea behind CNNs came from recognizing that low-level image statistics appear everywhere in an image and are not necessarily limited to a particular location.
Additional resources:
