Menu
Home
People
Places
Arts
History
Plants & Animals
Science
Life & Culture
Technology
Reference.org
Multilayer perceptron
open-in-new
<p>In <a href="/facts/Deep_learning/JLuwD3ea">deep learning</a>, a multilayer perceptron (MLP) is a name for a modern <a href="/facts/Feedforward_neural_network/CP0pPGDF">feedforward</a> <a href="/facts/Neural_network_(machine_learning)/6V1jMlkx">neural network</a> consisting of fully connected neurons with nonlinear <a href="/facts/Activation_function/S4NImL6L">activation functions</a>, organized in layers, notable for being able to distinguish data that is not <a href="/facts/Linear_separability/TNES6fgy">linearly separable</a>. </p><p>Modern neural networks are trained using <a href="/facts/Backpropagation/lCsIdKHc">backpropagation</a> and are colloquially referred to as "vanilla" networks. MLPs grew out of an effort to improve <a href="/facts/Perceptron/ArxdkAC1">single-layer perceptrons</a>, which could only be applied to linearly separable data. A perceptron traditionally used a <a href="/facts/Heaviside_step_function/bgsUtxgP">Heaviside step function</a> as its nonlinear activation function. However, the backpropagation algorithm requires that modern MLPs use <a href="/facts/Continuous_function/sKbl02pB">continuous</a> activation functions such as <a href="/facts/Sigmoid_function/lAXu3AwW">sigmoid</a> or <a href="/facts/Rectifier_(neural_networks)/pwzdU9jF">ReLU</a>. </p><p>Multilayer perceptrons form the basis of deep learning, and are <a href="/facts/Neural_network_(machine_learning)/6V1jMlkx">applicable</a> across a vast set of diverse domains. </p>