This is a type of feed-forward neural network with multiple layers that often uses a supervised learning algorithm. This type of network contains a number of adaptive nodes that are interconnected, without any weight value between them. Each node in this network has different functions and tasks. A learning rule that is used affects parameters in the node and reduces error levels at the output layer.
This neural network is usually trained with backpropagation or gradient descent. Due to the slowness in convergence, a hybrid approach can also be used, which accelerates the convergence and potentially avoids local minima.
