|
gptkbp:instance_of
|
gptkb:technique
|
|
gptkbp:applies_to
|
deep learning
|
|
gptkbp:can_be_applied_during
|
training phase
inference phase
|
|
gptkbp:can_be_combined_with
|
residual networks
|
|
gptkbp:can_be_used_with
|
dropout
|
|
gptkbp:can_lead_to
|
better generalization
|
|
gptkbp:can_provide
|
overfitting
|
|
gptkbp:developed_by
|
gptkb:Christian_Szegedy
gptkb:Sergey_Ioffe
|
|
gptkbp:enhances
|
gradient flow
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
Batch Normalization
|
|
gptkbp:improves
|
training speed
|
|
gptkbp:includes
|
scaling factor
bias term
|
|
gptkbp:introduced_in
|
gptkb:2015
|
|
gptkbp:is_a_form_of
|
layer normalization
|
|
gptkbp:is_a_key_component_of
|
modern architectures
|
|
gptkbp:is_applied_in
|
fully connected layers
recurrent neural networks
convolutional layers
sequence models
|
|
gptkbp:is_beneficial_for
|
very deep networks
|
|
gptkbp:is_implemented_in
|
gptkb:Tensor_Flow
gptkb:Keras
gptkb:Py_Torch
|
|
gptkbp:is_often_accompanied_by
|
gptkb:neural_networks
|
|
gptkbp:is_often_used_in
|
gptkb:generative_adversarial_networks
convolutional neural networks
|
|
gptkbp:is_part_of
|
gptkb:neural_networks
|
|
gptkbp:is_related_to
|
group normalization
instance normalization
|
|
gptkbp:is_used_in
|
natural language processing
image classification
object detection
transfer learning
|
|
gptkbp:is_used_to_mitigate
|
vanishing gradients
|
|
gptkbp:normalizes
|
layer inputs
|
|
gptkbp:reduces
|
internal covariate shift
|
|
gptkbp:requires
|
computational overhead
|
|
gptkbp:sensitivity
|
batch size
|
|
gptkbp:technique
|
accelerating convergence
improves model stability
normalizes the output of a previous layer.
normalizing activations
reduces sensitivity to initialization
|
|
gptkbp:uses
|
mini-batch statistics
|
|
gptkbp:bfsParent
|
gptkb:neural_networks
gptkb:Deep_Learning
gptkb:Sergey_Ioffe
|
|
gptkbp:bfsLayer
|
4
|