Class TripletLoss<T>

Namespace

AiDotNet.LossFunctions

Assembly

AiDotNet.dll

Implements the Triplet Loss function for learning similarity embeddings.

public class TripletLoss<T> : LossFunctionBase<T>, ILossFunction<T>

Type Parameters

T

The numeric type used for calculations (e.g., float, double).

Inheritance

object

LossFunctionBase<T>

TripletLoss<T>

Implements

ILossFunction<T>

Inherited Members

LossFunctionBase<T>.NumOps

LossFunctionBase<T>.Engine

LossFunctionBase<T>.CalculateLossAndGradientGpu(IGpuTensor<T>, IGpuTensor<T>)

LossFunctionBase<T>.ValidateVectorLengths(Vector<T>, Vector<T>)

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Extension Methods

LossFunctionExtensions.ComputeGradient<T>(ILossFunction<T>, Tensor<T>, Tensor<T>)

LossFunctionExtensions.ComputeLoss<T>(ILossFunction<T>, Tensor<T>, Tensor<T>)

Remarks

For Beginners: Triplet Loss helps create embeddings (numerical representations) where similar items are close together and different items are far apart in a vector space.

It works with triplets of data:

• Anchor: A reference point (e.g., a person's face)
• Positive: An example similar to the anchor (e.g., another image of the same person)
• Negative: An example different from the anchor (e.g., an image of a different person)

The loss encourages the model to make the distance between the anchor and positive smaller than the distance between the anchor and negative by at least a specified margin.

This loss function is commonly used in:

• Face recognition systems
• Image retrieval applications
• Recommendation systems
• Any task where you need to learn meaningful similarity metrics

By minimizing triplet loss, the model learns to create an embedding space where semantically similar items cluster together and dissimilar items are pushed apart.

Constructors

TripletLoss(double)

Initializes a new instance of the TripletLoss class.

public TripletLoss(double margin = 1)

Parameters

margin double

The minimum desired difference between positive and negative distances. Default is 1.0.

Methods

CalculateDerivative(Matrix<T>, Matrix<T>, Matrix<T>)

Calculates the gradients of the Triplet Loss function for anchor, positive, and negative samples.

public (Matrix<T>, Matrix<T>, Matrix<T>) CalculateDerivative(Matrix<T> anchor, Matrix<T> positive, Matrix<T> negative)

Parameters

anchor Matrix<T>

The anchor samples matrix.

positive Matrix<T>

The positive samples matrix (similar to anchor).

negative Matrix<T>

The negative samples matrix (dissimilar to anchor).

Returns

(Matrix<T> U, Matrix<T> B, Matrix<T> V)

A tuple containing the gradients for anchor, positive, and negative samples.

Exceptions

ArgumentException

Thrown when input matrices have inconsistent dimensions.

CalculateDerivative(Vector<T>, Vector<T>)

This method is not used for Triplet Loss as it requires multiple input vectors.

public override Vector<T> CalculateDerivative(Vector<T> predicted, Vector<T> actual)

Parameters

predicted Vector<T>

The predicted values vector.

actual Vector<T>

The actual (target) values vector.

Returns

Vector<T>

Throws NotSupportedException.

Exceptions

NotSupportedException

Always thrown as TripletLoss requires three input matrices.

CalculateLoss(Matrix<T>, Matrix<T>, Matrix<T>)

Calculates the Triplet Loss for embedding learning.

public T CalculateLoss(Matrix<T> anchor, Matrix<T> positive, Matrix<T> negative)

Parameters

anchor Matrix<T>

The anchor samples (reference points).

positive Matrix<T>

The positive samples (similar to anchors).

negative Matrix<T>

The negative samples (dissimilar to anchors).

Returns

T

A scalar value representing the triplet loss.

Exceptions

ArgumentException

Thrown when input matrices have inconsistent dimensions.

CalculateLoss(Vector<T>, Vector<T>)

This method is not used for Triplet Loss as it requires multiple input vectors.

public override T CalculateLoss(Vector<T> predicted, Vector<T> actual)

Parameters

predicted Vector<T>

The predicted values vector.

actual Vector<T>

The actual (target) values vector.

Returns

T

Throws NotSupportedException.

Exceptions

NotSupportedException

Always thrown as TripletLoss requires three input matrices.

CalculateLossAndGradientGpu(IGpuTensor<T>, IGpuTensor<T>, IGpuTensor<T>)

Calculates Triplet Loss on GPU for batched input tensors.

public (T Loss, IGpuTensor<T> AnchorGradient, IGpuTensor<T> PositiveGradient, IGpuTensor<T> NegativeGradient) CalculateLossAndGradientGpu(IGpuTensor<T> anchor, IGpuTensor<T> positive, IGpuTensor<T> negative)

Parameters

anchor IGpuTensor<T>

The anchor GPU tensor (batch of embeddings).

positive IGpuTensor<T>

The positive GPU tensor (similar to anchors).

negative IGpuTensor<T>

The negative GPU tensor (dissimilar to anchors).

Returns

(T Loss, IGpuTensor<T> AnchorGradient, IGpuTensor<T> PositiveGradient, IGpuTensor<T> NegativeGradient)

A tuple containing the loss value and gradient tensors for anchor, positive, and negative.