Loss Functions

Complete reference for all 41 loss functions in AiDotNet.

Classification Losses

Binary Classification

Loss Function	Description	Use Case
`BinaryCrossEntropyLoss<T>`	-[y·log(p) + (1-y)·log(1-p)]	Binary classification
`HingeLoss<T>`	max(0, 1 - y·p)	SVM-style binary classification
`SquaredHingeLoss<T>`	(max(0, 1 - y·p))^2	Smooth differentiable hinge
`ModifiedHuberLoss<T>`	Modified hinge + quadratic	Robust binary

var loss = new BinaryCrossEntropyLoss<float>();
var value = loss.CalculateLoss(predictions, targets);

Multi-Class Classification

Loss Function	Description	Use Case
`CrossEntropyLoss<T>`	Standard cross entropy	Multi-class
`CategoricalCrossEntropyLoss<T>`	One-hot encoded cross entropy	Multi-class with one-hot
`SparseCategoricalCrossEntropyLoss<T>`	Integer label cross entropy	Multi-class with indices
`WeightedCrossEntropyLoss<T>`	Class-weighted cross entropy	Imbalanced classes

var loss = new CrossEntropyLoss<float>();

Focal Loss (Imbalanced Data)

Loss Function	Description	Use Case
`FocalLoss<T>`	Down-weights easy examples	Class imbalance

var loss = new FocalLoss<float>();

Regression Losses

Loss Function	Description	Use Case
`MeanSquaredErrorLoss<T>`	Mean squared error: (y - p)^2	General regression
`MeanAbsoluteErrorLoss<T>`	Mean absolute error	Robust to outliers
`RootMeanSquaredErrorLoss<T>`	Square root of MSE	Same scale as target
`MeanBiasErrorLoss<T>`	Mean bias error	Directional bias
`HuberLoss<T>`	Quadratic near 0, linear elsewhere	Balanced robustness
`LogCoshLoss<T>`	log(cosh(y - p))	Smooth L1 approximation
`CharbonnierLoss<T>`	Differentiable L1	Image restoration
`PoissonLoss<T>`	Poisson negative log likelihood	Count data
`QuantileLoss<T>`	Quantile regression	Prediction intervals

var loss = new HuberLoss<float>();

Segmentation Losses

Loss Function	Description	Use Case
`DiceLoss<T>`	1 - Dice coefficient	Segmentation
`JaccardLoss<T>`	1 - Intersection over Union	Object detection
`ScaleInvariantDepthLoss<T>`	Scale-invariant depth	Depth estimation

var loss = new DiceLoss<float>();

Contrastive/Metric Learning

Loss Function	Description	Use Case
`ContrastiveLoss<T>`	Siamese networks	Similarity learning
`TripletLoss<T>`	Anchor, positive, negative	Face recognition
`NTXentLoss<T>`	Normalized temperature cross-entropy	Self-supervised
`InfoNCELoss<T>`	Information NCE	Contrastive learning
`NoiseContrastiveEstimationLoss<T>`	NCE for large vocabularies	Word embeddings
`CosineSimilarityLoss<T>`	1 - cosine similarity	Embedding alignment

var loss = new TripletLoss<float>();

Reconstruction Losses

Loss Function	Description	Use Case
`MAEReconstructionLoss<T>`	MAE for reconstruction	Masked autoencoders
`PerceptualLoss<T>`	Feature space distance	Super-resolution
`RealESRGANLoss<T>`	Combined perceptual + adversarial	Image enhancement
`RotationPredictionLoss<T>`	Rotation prediction	Self-supervised pretext

var loss = new PerceptualLoss<float>();

GAN Losses

Loss Function	Description	Use Case
`WassersteinLoss<T>`	Wasserstein distance	WGAN
`MarginLoss<T>`	Margin-based loss	Pairwise ranking

var loss = new WassersteinLoss<float>();

Other Losses

Loss Function	Description	Use Case
`ElasticNetLoss<T>`	L1 + L2 combined	Regularized regression
`ExponentialLoss<T>`	Exponential loss	AdaBoost
`OrdinalRegressionLoss<T>`	Ordered categories	Ordinal prediction
`QuantumLoss<T>`	Quantum-inspired loss	Quantum ML

Sequence Losses

Loss Function	Description	Use Case
`CTCLoss<T>`	Connectionist Temporal Classification	Speech recognition

var loss = new CTCLoss<float>();

Self-Supervised Learning Losses

Loss Function	Description	Use Case
`BYOLLoss<T>`	Bootstrap Your Own Latent	Self-supervised
`BarlowTwinsLoss<T>`	Redundancy reduction	Self-supervised
`DINOLoss<T>`	Self-distillation	Vision transformers

Usage Examples

With AiModelBuilder

var result = await new AiModelBuilder<float, float[][], float[]>()
    .ConfigureModel(model)
    .ConfigureLossFunction(new FocalLoss<float>())
    .BuildAsync(trainData, trainLabels);

Loss Selection Guide

Task	Recommended Loss
Binary classification	`BinaryCrossEntropyLoss`
Multi-class classification	`CrossEntropyLoss`
Imbalanced classification	`FocalLoss`
Regression	`MeanSquaredErrorLoss` or `HuberLoss`
Segmentation	`DiceLoss` + `CrossEntropyLoss`
Object detection	`JaccardLoss`
Face recognition	`TripletLoss`
Contrastive learning	`NTXentLoss` or `InfoNCELoss`
GANs	`WassersteinLoss`
Speech recognition	`CTCLoss`
Self-supervised	`BYOLLoss` or `DINOLoss`