<!--
{
  "availability" : [
    "iOS: -",
    "iPadOS: -",
    "macCatalyst: -",
    "macOS: -",
    "tvOS: -",
    "visionOS: -",
    "watchOS: -"
  ],
  "documentType" : "symbol",
  "framework" : "Accelerate",
  "identifier" : "/documentation/Accelerate/BNNSLayerParametersReduction/init(i_desc:o_desc:w_desc:reduce_func:epsilon:)",
  "metadataVersion" : "0.1.0",
  "role" : "Initializer",
  "symbol" : {
    "kind" : "Initializer",
    "modules" : [
      "Accelerate"
    ],
    "preciseIdentifier" : "s:So28BNNSLayerParametersReductiona6i_desc02o_D002w_D011reduce_func7epsilonABSo21BNNSNDArrayDescriptora_A2ISo18BNNSReduceFunctionaSftcfc"
  },
  "title" : "init(i_desc:o_desc:w_desc:reduce_func:epsilon:)"
}
-->

# init(i_desc:o_desc:w_desc:reduce_func:epsilon:)

Returns a structure containing the parameters of a reduction layer from the specified parameters.

```
init(i_desc: BNNSNDArrayDescriptor, o_desc: BNNSNDArrayDescriptor, w_desc: BNNSNDArrayDescriptor, reduce_func: BNNSReduceFunction, epsilon: Float)
```

## Parameters

`i_desc`

The descriptor of the input.

`o_desc`

The descriptor of the output.

`w_desc`

The descriptor of the weights.

`reduce_func`

The variable that specifies the reduction function.

`epsilon`

A value that the operation adds to each element when computing the sum of logarithms.

## Discussion

> Important:
> The number of input dimensions must be equal to number of output dimensions, and equal to the number of weights dimensions. The reduction layer only supports `float`, with the exception of ``doc://com.apple.accelerate/documentation/Accelerate/BNNSReduceFunctionLogicalOr`` and ``doc://com.apple.accelerate/documentation/Accelerate/BNNSReduceFunctionLogicalAnd`` that support `float` and `BNNSDataTypeBoolean`.

---

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)