Figure 1.

Anatomy of a feedforward artificial neural network. Illustration of a one-layer fully-connected network (FCN) architecture. The network consists of an input layer with three units, exemplified here by age, smoking status, and left ventricle ejection fraction. Hidden neurons receive inputs, compute weighted sums, and pass through nonlinear activation functions to produce outputs. The network maps process data to output units, providing probabilities of class membership or estimating numeric quantities of interest. The primary objective of this architecture is to learn complex nonlinear mappings between input and ground-truth data.