Piecewise Linear Sigmoid Approximation

In deep learning and neural network design, activation functions play a crucial role in introducing non-linearity into models. The sigmoid function is one of the classic activation functions, but its computation involves an expensive exponential operation. To address this computational overhead, practitioners often use the Piecewise Linear Sigmoid Approximation (sometimes called the "hard sigmoid")—a fast, hardware-friendly alternative.

The Piecewise Linear Sigmoid Approximation replaces the smooth S-curve of the standard sigmoid with a simple linear function in the active region and hard saturation at the boundaries. This makes it highly efficient for embedded systems, mobile devices, and scenarios where computational resources are limited.

Mathematical Definition:

The piecewise linear sigmoid is defined as:

$$ \text{piecewise_sigmoid}(x) = \begin{cases} 0 & \text{if } x \leq -2.5 \ 0.2x + 0.5 & \text{if } -2.5 < x < 2.5 \ 1 & \text{if } x \geq 2.5 \end{cases} $$

Key Characteristics:

• Saturation Regions: For inputs less than or equal to -2.5, the output is clamped to 0. For inputs greater than or equal to 2.5, the output is clamped to 1.
• Linear Region: In the range (-2.5, 2.5), the function behaves linearly with slope 0.2 and intercept 0.5.
• Bounded Output: The output is always in the range [0, 1], making it suitable for probability-like activations.
• Continuous Function: The function is continuous everywhere, though not differentiable at the boundary points.

Your Task:

Implement a function that computes the piecewise linear sigmoid approximation for a given input value. The function should correctly handle all three regions: the left saturation zone, the linear transition zone, and the right saturation zone.