Class FusedConvBatchNormOp

Namespace

AiDotNet.JitCompiler.IR.Operations

Assembly

AiDotNet.dll

Fused convolution + batch normalization operation.

public class FusedConvBatchNormOp : IROp

Inheritance

object

IROp

FusedConvBatchNormOp

Inherited Members

IROp.OutputIds

IROp.OutputId

IROp.InputIds

IROp.OutputType

IROp.OutputShape

IROp.OutputShapes

IROp.OpType

IROp.ToString()

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

Remarks

For Beginners: Combines convolution with batch normalization.

Batch normalization after convolution is extremely common in CNNs. By fusing them, we can:

• Fold BN parameters into conv weights (at inference time)
• Skip intermediate tensor storage
• Reduce memory bandwidth significantly

This can be 2-3x faster than separate operations!

Properties

Epsilon

Gets or sets the batch norm epsilon value.

public double Epsilon { get; set; }

Property Value

double

Momentum

Gets or sets the batch norm momentum.

public double Momentum { get; set; }

Property Value

double

Padding

Gets or sets the convolution padding.

public int[] Padding { get; set; }

Property Value

int[]

Stride

Gets or sets the convolution stride.

public int[] Stride { get; set; }

Property Value

int[]

Methods

Validate()

Validates inputs (input, kernel, gamma, beta, running_mean, running_var).

public override bool Validate()

Returns

bool