A neural network layer that outputs the maximum value across a set of linear functions, enabling piecewise linear approximations.