SquashActivation<T>

Models & Types · AiDotNet.ActivationFunctions

Implements the Squash activation function, which normalizes vectors to have a magnitude between 0 and 1.

Inheritance: ActivationFunctionBase<T> → SquashActivation<T>

For Beginners

The Squash function is different from most activation functions because it works on entire vectors (groups of numbers) rather than individual numbers. Think of a vector as an arrow pointing in some direction with a certain length.

What Squash does:

  1. It keeps the arrow pointing in the same direction
  2. It adjusts the length of the arrow to be between 0 and 1
  3. Short vectors remain almost the same length
  4. Long vectors get “squashed” to be closer to length 1

This is useful in advanced neural networks like capsule networks, where the direction of a vector represents important features, and the length represents the probability that those features exist.

How It Works

The Squash activation function maps input vectors to have a magnitude (length) between 0 and 1 while preserving their direction. This is particularly useful in capsule networks and other neural network architectures where vector orientation is important.

Example

using AiDotNet.ActivationFunctions;

var activation = new SquashActivation<double>();
double y = activation.Activate(0.5);
Console.WriteLine($"SquashActivation: f(0.5) = {y:F4}");

Constructors

Constructor Summary
SquashActivation()

Methods

Method Summary
Matrix<T> Derivative(Vector<T> input) Calculates the Jacobian matrices for a batch of vectors stored in a tensor.
T Activate(T input) Applies the Squash activation function to a batch of vectors stored in a tensor.
T Derivative(T input) Calculates the Jacobian matrices for a batch of vectors stored in a tensor.
Tensor<T> Activate(Tensor<T> input) Applies the Squash activation function to a batch of vectors stored in a tensor.
Tensor<T> Derivative(Tensor<T> input) Calculates the Jacobian matrices for a batch of vectors stored in a tensor.
Vector<T> Activate(Vector<T> input) Applies the Squash activation function to a batch of vectors stored in a tensor.