Multilayer Perceptron (MLP) or Feed Forward Network (FFN) is usually the first type of neuron network that one comes to understand before graduating to Restricted Boltzmann Machines (RBM) and Recurrent Neural Network (RRN) that are common in deep learning. While MLP technically can be configured/referred to as deep network, one must investigate a bit and understand as to why it is considered a first step (only) in a journey toward deep learning networks.
In Spark's 2.0 implementation, sigmoid function (non-linear activation) is used in a deep stackable network configuration (more than three layer) to map the output to a Softmax function to create a non-trivial mapping surface that can capture extreme non-linear behavior of ...
