ai.djl.nn.norm

Class LayerNorm

java.lang.Object

ai.djl.nn.AbstractBlock

ai.djl.nn.norm.LayerNorm

All Implemented Interfaces:

Block

public class LayerNorm
extends AbstractBlock

Layer normalization works by normalizing the values of input data for each input sample to have mean of 0 and variance of 1. Since this may alter the representation of a layer, two parameters (\ (\gamma\) and \(\beta\)) are learned along the normalization process to respectively scale and shift the normalized output (activations) to have any mean and variance so the network can utilize non-linear transformations such as sigmoid function as described in the paper. During backpropagation, both \(\gamma\) and \(\beta\) parameters are included following the chain-rule in derivation.

Citing the abstract of the paper: "Training state-of-the-art, deep neural networks is computationally expensive. One way to reduce the training time is to normalize the activities of the neurons. A recently introduced technique called batch normalization uses the distribution of the summed input to a neuron over a mini-batch of training cases to compute a mean and variance which are then used to normalize the summed input to that neuron on each training case. This significantly reduces the training time in feed-forward neural networks. However, the effect of batch normalization is dependent on the mini-batch size and it is not obvious how to apply it to recurrent neural networks. In this paper, we transpose batch normalization into layer normalization by computing the mean and variance used for normalization from all of the summed inputs to the neurons in a layer on a single training case. Like batch normalization, we also give each neuron its own adaptive bias and gain which are applied after the normalization but before the non-linearity. Unlike batch normalization, layer normalization performs exactly the same computation at training and test times. It is also straightforward to apply to recurrent neural networks by computing the normalization statistics separately at each time step. Layer normalization is very effective at stabilizing the hidden state dynamics in recurrent networks. Empirically, we show that layer normalization can substantially reduce the training time compared with previously published techniques."

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LayerNorm.Builder The Builder to construct a LayerNorm.

Field Summary

Fields inherited from class ai.djl.nn.AbstractBlock

children, inputNames, inputShapes, parameters, version

Method Summary

Modifier and Type	Method and Description
protected void	beforeInitialize(Shape... inputShapes) Performs any action necessary before initialization.
static LayerNorm.Builder	builder() Creates a builder to build a LayerNorm.
protected NDList	forwardInternal(ParameterStore parameterStore, NDList inputs, boolean training, ai.djl.util.PairList<java.lang.String,java.lang.Object> params) A helper for Block.forward(ParameterStore, NDList, boolean, PairList) after initialization.
Shape[]	getOutputShapes(Shape[] inputShapes) Returns the expected output shapes of the block for the specified input shapes.
static NDList	layerNorm(NDArray input, Shape normalizedShape, NDArray gamma, NDArray beta, float eps) Applies Layer Normalization with average and variance for each input sample across the axis dimensions.
void	loadMetadata(byte loadVersion, java.io.DataInputStream is) Overwrite this to load additional metadata with the parameter values.
void	prepare(Shape[] inputShapes) Sets the shape of Parameters.
protected void	saveMetadata(java.io.DataOutputStream os) Override this method to save additional data apart from parameter values.

Methods inherited from class ai.djl.nn.AbstractBlock

addChildBlock, addParameter, cast, clear, describeInput, forward, forward, forwardInternal, getChildren, getDirectParameters, getParameters, initialize, initializeChildBlocks, isInitialized, loadParameters, readInputShapes, saveInputShapes, saveParameters, setInitializer, setInitializer, setInitializer, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface ai.djl.nn.Block

forward, validateLayout

Method Detail

layerNorm

public static NDList layerNorm(NDArray input,
                               Shape normalizedShape,
                               NDArray gamma,
                               NDArray beta,
                               float eps)

Applies Layer Normalization with average and variance for each input sample across the axis dimensions.

Parameters:

input - the input NDArray of shape (batchSize, inputChannel, *), * could be empty, width, (height, width), (depth, height, width)

normalizedShape - dimensions to calculate average and variance from

gamma - gamma weight NDArray

beta - beta weight NDArray

eps - a value added to the denominator for numerical stability

Returns:

the output NDArray of shape (batchSize, inputChannel, *), * could be empty, width, (height, width), (depth, height, width)

builder

public static LayerNorm.Builder builder()

Creates a builder to build a LayerNorm.

Returns:

a new builder

forwardInternal

protected NDList forwardInternal(ParameterStore parameterStore,
                                 NDList inputs,
                                 boolean training,
                                 ai.djl.util.PairList<java.lang.String,java.lang.Object> params)

A helper for Block.forward(ParameterStore, NDList, boolean, PairList) after initialization.

Specified by:

forwardInternal in class AbstractBlock

Parameters:

parameterStore - the parameter store

inputs - the input NDList

training - true for a training forward pass

params - optional parameters

Returns:

the output of the forward pass

getOutputShapes

public Shape[] getOutputShapes(Shape[] inputShapes)

Returns the expected output shapes of the block for the specified input shapes.

Parameters:

inputShapes - the shapes of the inputs

Returns:

the expected output shapes of the block

beforeInitialize

protected void beforeInitialize(Shape... inputShapes)

Performs any action necessary before initialization. For example, keep the input information or verify the layout.

Overrides:

beforeInitialize in class AbstractBlock

Parameters:

inputShapes - the expected shapes of the input

prepare

public void prepare(Shape[] inputShapes)

Sets the shape of Parameters.

Overrides:

prepare in class AbstractBlock

Parameters:

inputShapes - the shapes of inputs

saveMetadata

protected void saveMetadata(java.io.DataOutputStream os)
                     throws java.io.IOException

Override this method to save additional data apart from parameter values.

This default implementation saves the currently set input shapes.

Overrides:

saveMetadata in class AbstractBlock

Parameters:

os - the non-null output stream the parameter values and metadata are written to

Throws:

java.io.IOException - saving failed

loadMetadata

public void loadMetadata(byte loadVersion,
                         java.io.DataInputStream is)
                  throws java.io.IOException,
                         MalformedModelException

Overwrite this to load additional metadata with the parameter values.

If you overwrite AbstractBlock.saveMetadata(DataOutputStream) or need to provide backward compatibility to older binary formats, you prabably need to overwrite this. This default implementation checks if the version number fits, if not it throws an MalformedModelException. After that it restores the input shapes.

Overrides:

loadMetadata in class AbstractBlock

Parameters:

loadVersion - the version used for loading this metadata.

is - the input stream we are loading from

Throws:

java.io.IOException - loading failed

MalformedModelException - data can be loaded but has wrong format