The input layer holds the image data. In the following figure, the input layer consists of three inputs. In a fully connected layer, the neurons between two adjacent layers are fully connected pairwise but do not share any connection within a layer. In other words, the neurons in this layer have full connections to all activations in the previous layer. Therefore, their activations can be computed with a simple matrix multiplication, optionally adding a bias term. The difference between a fully connected and convolutional layer is that neurons in a convolutional layer are connected to a local region in the input, and that they also share parameters: