---
title: "applyBackward(batchSize:inputA:inputB:output:outputGradient:generatingInputAGradient:generatingInputBGradient:generatingParameterGradients:)"
framework: accelerate
role: symbol
role_heading: Instance Method
path: "accelerate/bnns/fusedparameterslayer/applybackward(batchsize:inputa:inputb:output:outputgradient:generatinginputagradient:generatinginputbgradient:generatingparametergradients:)"
---

# applyBackward(batchSize:inputA:inputB:output:outputGradient:generatingInputAGradient:generatingInputBGradient:generatingParameterGradients:)

Applies the layer backward to generate input gradients, where the first layer accepts two inputs.

## Declaration

```swift
func applyBackward(batchSize: Int, inputA: BNNSNDArrayDescriptor, inputB: BNNSNDArrayDescriptor, output: BNNSNDArrayDescriptor, outputGradient: BNNSNDArrayDescriptor, generatingInputAGradient inputAGradient: BNNSNDArrayDescriptor, generatingInputBGradient inputBGradient: BNNSNDArrayDescriptor, generatingParameterGradients parameterGradients: [BNNSNDArrayDescriptor]) throws
```

## Parameters

- `batchSize`: The number of input-output pairs.
- `inputA`: The descriptor of the first input.
- `inputB`: The descriptor of the second input.
- `output`: The descriptor of the output.
- `outputGradient`: The descriptor of the output gradient.
- `inputAGradient`: The descriptor of the input gradient.
- `inputBGradient`: The descriptor of the input gradient.
- `parameterGradients`: An array that contains the parameter gradients.

## See Also

### Applying a Fused Parameters Layer

- [apply(batchSize:inputA:inputB:output:for:)](accelerate/bnns/fusedparameterslayer/apply(batchsize:inputa:inputb:output:for:).md)
- [apply(batchSize:inputA:inputB:inputC:output:for:)](accelerate/bnns/fusedparameterslayer/apply(batchsize:inputa:inputb:inputc:output:for:).md)
- [applyBackward(batchSize:inputA:inputB:inputC:output:outputGradient:generatingInputAGradient:generatingInputBGradient:generatingInputCGradient:generatingParameterGradients:)](accelerate/bnns/fusedparameterslayer/applybackward(batchsize:inputa:inputb:inputc:output:outputgradient:generatinginputagradient:generatinginputbgradient:generatinginputcgradient:generatingparametergradients:).md)