The MLP operates through multiple layers, including an input layer, one or more hidden layers, and an output layer. Each neuron in these layers, except for the input nodes, applies a nonlinear activation function, typically sigmoid, hyperbolic tangent, or ReLU, to its weighted input to produce an output. MLPs use a method known as backpropagation for learning, which involves adjusting the weights of the network based on the error rate obtained in the previous epoch (i.e., the difference between the actual output and the predicted output). This makes MLPs highly effective for complex pattern recognition tasks, such as speech recognition, image classification, and predictive analytics, by capturing non-linear relationships in data.
The concept of the perceptron, which is the building block of MLPs, was first introduced by Frank Rosenblatt in 1958. However, MLPs gained substantial popularity in the 1980s when the backpropagation algorithm was popularized, allowing these networks to be trained efficiently and used extensively in practical applications.
The development of MLPs is linked closely with the work of Frank Rosenblatt for the foundational perceptron model. The revival and popularization of MLPs in the 1980s were significantly influenced by David E. Rumelhart, Geoffrey Hinton, and Ronald J. Williams, who helped refine and popularize the backpropagation learning algorithm. Their contributions were crucial in demonstrating the practical applications and effectiveness of MLPs in various fields of artificial intelligence.