.TH "MPSNNGradientStateNode" 3 "Mon Jul 9 2018" "Version MetalPerformanceShaders-119.3" "MetalPerformanceShaders.framework" \" -*- nroff -*-
.ad l
.nh
.SH NAME
MPSNNGradientStateNode
.SH SYNOPSIS
.br
.PP
.PP
\fC#import <MPSNNGraphNodes\&.h>\fP
.PP
Inherits \fBMPSNNStateNode\fP\&.
.PP
Inherited by \fBMPSCNNConvolutionGradientStateNode\fP\&.
.SS "Additional Inherited Members"
.SH "Detailed Description"
.PP 
During training, each \fBMPSNNFilterNode\fP has a corresponding \fBMPSNNGradientFilterNode\fP for the gradient computation for trainable parameter update\&. The two communicate through a \fBMPSNNGradientStateNode\fP or subclass which carries information about the inference pass settings to the gradient pass\&. You can avoid managing these -- there will be many! -- by using -[\fBMPSNNFilterNode\fP gradientFilterWithSources:] to make the MPSNNGradientFilterNodes\&. That method will append the necessary extra information like \fBMPSNNGradientState\fP nodes and inference filter source image nodes to the object as needed\&. 

.SH "Author"
.PP 
Generated automatically by Doxygen for MetalPerformanceShaders\&.framework from the source code\&.