dl::activation::Softmax Class Reference

Softmax activation function. More...

#include <functions.hpp>

Inheritance diagram for dl::activation::Softmax:

Collaboration diagram for dl::activation::Softmax:

Public Member Functions
MatrixD	forward (const MatrixD &x)
	Compute softmax forward pass.
MatrixD	backward (const MatrixD &x)
	Compute softmax Jacobian matrix.
Public Member Functions inherited from dl::activation::ActivationFunction
virtual	~ActivationFunction ()=default
	Virtual destructor for proper cleanup.

Detailed Description

Softmax activation function.

Softmax is commonly used in the output layer of multi-class classification networks. It converts a vector of real numbers into a probability distribution.

Mathematical definition:

• Forward: f(x_i) = exp(x_i) / Σ(exp(x_j)) for j=1 to n
• Backward: Jacobian matrix with elements ∂f_i/∂x_j

Note

Output values sum to 1, making them interpretable as probabilities

Warning

Numerically unstable for large input values without proper scaling

Definition at line 140 of file functions.hpp.

Member Function Documentation

backward()

MatrixD dl::activation::Softmax::backward

(

const MatrixD &

x

)

Compute softmax Jacobian matrix.

Parameters

x	Input matrix/vector

Returns

Jacobian matrix for backpropagation

Definition at line 55 of file functions.cpp.

forward()

MatrixD dl::activation::Softmax::forward

(

const MatrixD &

x

)

Compute softmax forward pass.

Parameters

x	Input matrix/vector

Returns

Probability distribution matrix

Definition at line 48 of file functions.cpp.

---

The documentation for this class was generated from the following files:

• include/activation/functions.hpp
• src/activation/functions.cpp