<!--
{
  "availability" : [
    "*: -",
    "iOS: 14.0.0 -",
    "iPadOS: 14.0.0 -",
    "macCatalyst: -",
    "macOS: 11.0.0 -",
    "tvOS: 14.0.0 -",
    "visionOS: -",
    "watchOS: 7.0.0 -"
  ],
  "documentType" : "symbol",
  "framework" : "Accelerate",
  "identifier" : "/documentation/Accelerate/BNNS/LossFunction",
  "metadataVersion" : "0.1.0",
  "role" : "Enumeration",
  "symbol" : {
    "kind" : "Enumeration",
    "modules" : [
      "Accelerate"
    ],
    "preciseIdentifier" : "s:10Accelerate4BNNSO12LossFunctionO"
  },
  "title" : "BNNS.LossFunction"
}
-->

# BNNS.LossFunction

Constants that describe loss functions.

```
enum LossFunction
```

## Topics

### Loss Functions

[`case categoricalCrossEntropy`](/documentation/Accelerate/BNNS/LossFunction/categoricalCrossEntropy)

Categorical cross-entropy computation between input prediction and labels.

[`case cosineDistance`](/documentation/Accelerate/BNNS/LossFunction/cosineDistance)

Cosine distance loss computation between input predictions and labels.

[`case hinge`](/documentation/Accelerate/BNNS/LossFunction/hinge)

Hinge loss computation between labels and unbounded, zero-centered binary predictions.

[`case huber(huberDelta: Float)`](/documentation/Accelerate/BNNS/LossFunction/huber(huberDelta:))

Huber loss computation between input logits and one-hot encoded labels.

[`case log`](/documentation/Accelerate/BNNS/LossFunction/log)

Log loss computation between labels and predictions.

[`case meanAbsoluteError`](/documentation/Accelerate/BNNS/LossFunction/meanAbsoluteError)

Mean absolute error (MAE) computation between input prediction and labels.

[`case meanSquareError`](/documentation/Accelerate/BNNS/LossFunction/meanSquareError)

Mean square error (MSE) computation between input logits and one-hot encoded labels.

[`case sigmoidCrossEntropy(labelSmoothing: Float)`](/documentation/Accelerate/BNNS/LossFunction/sigmoidCrossEntropy(labelSmoothing:))

Sigmoid activation on input logits, and independent computation of cross-entropy loss for each class.

[`case softmaxCrossEntropy(labelSmoothing: Float)`](/documentation/Accelerate/BNNS/LossFunction/softmaxCrossEntropy(labelSmoothing:))

Softmax activation on input logits, and computation of cross-entropy loss with one-hot encoded labels.

[`case yolo(parameters: BNNS.LossFunction.YoloParameters)`](/documentation/Accelerate/BNNS/LossFunction/yolo(parameters:))

You Only Look Once (YOLO) loss computation between prediction and ground truth labels.

[`struct YoloParameters`](/documentation/Accelerate/BNNS/LossFunction/YoloParameters)

A structure that contains the parameters for You Only Look Once (YOLO) loss computation.

### Instance Properties

[`var bnnsLossFunction: BNNSLossFunction`](/documentation/Accelerate/BNNS/LossFunction/bnnsLossFunction)

The underlying loss function structure.


---

Copyright &copy; 2026 Apple Inc. All rights reserved. | [Terms of Use](https://www.apple.com/legal/internet-services/terms/site.html) | [Privacy Policy](https://www.apple.com/privacy/privacy-policy)