Skip to contents

The hard sigmoid activation is defined as:

  • 0 if if x < -2.5

  • 1 if x > 2.5

  • 0.2 * x + 0.5 if -2.5 <= x <= 2.5

It's a faster, piecewise linear approximation of the sigmoid activation.

Usage

activation_hard_sigmoid(x)

Arguments

x

Input tensor.

Value

A tensor, the result from applying the activation to the input tensor x.