MarginLoss<T>

Models & Types ยท AiDotNet.LossFunctions

Implements the Margin loss function, specifically designed for Capsule Networks.

Inheritance: LossFunctionBase<T> โ†’ MarginLoss<T>

For Beginners

Margin loss is a special loss function used in Capsule Networks.

The formula is: T_c * max(0, m+ - ||v_c||)^2 + lambda * (1 - T_c) * max(0, ||v_c|| - m-)^2

Where:

  • T_c is 1 if class c is present, 0 otherwise
  • ||v_c|| is the length of the output vector of the capsule for class c
  • m+ is the upper bound (usually 0.9)
  • m- is the lower bound (usually 0.1)
  • lambda is a down-weighting factor (usually 0.5)

Key properties:

  • Encourages the network to output high values for correct classes
  • Discourages high outputs for incorrect classes
  • Helps in learning to represent different aspects of the input

Margin loss is ideal for Capsule Networks because:

  • It allows multiple classes to be present in the same image
  • It encourages the network to learn to represent different viewpoints and transformations
  • It helps in achieving equivariance, a key property of Capsule Networks

Example

using AiDotNet.LossFunctions;
using AiDotNet.Tensors.LinearAlgebra;

var loss = new MarginLoss<float>();
var predicted = new Vector<float>(new[] { 0.9f, 0.2f, 0.7f });
var actual = new Vector<float>(new[] { 1.0f, 0.0f, 1.0f });

float value = loss.CalculateLoss(predicted, actual);
Console.WriteLine($"MarginLoss = {value:F4}");

Constructors

Constructor Summary
MarginLoss(double mPlus, double mMinus, double lambda) Initializes a new instance of the MarginLoss class with the specified parameters.

Methods

Method Summary
T CalculateLoss(Vector<T> predicted, Vector<T> actual) Calculates the Margin loss between predicted and actual values.
Tensor<T> ComputeTapeLoss(Tensor<T> predicted, Tensor<T> target)