'importTensorFlowLayers' does not support 'tf.keras.​layers.Nor​malization​'

Hi,

I am using the latest relase version of Matlab, and I think 'importTensorFlowLayer's does not support 'tf.keras.layers.Normalization'.

https://www.tensorflow.org/api_docs/python/tf/keras/layers/Normalization

'tf.keras.layers.Normalization' layer remembers mean and variance in its layer to nomralize input data. Do you have a plan to support this layer, or how can I replace this layer to work in Matlab?

Thanks.

Answer 1

[Bhuvanyu Dogra]

It is my understanding that you are looking for alternatives of the layer 'tf.keras.layers.Normalization' in MATLAB R2021a.

'layerNormalizationLayer' or 'batchNormalizationLayer' can serve as alternatives as they provide normalization across all channels or individual channels. Moreover, 'batchNormalizationLayer' has the ability to use mean or variance for prediction.

For more information, refer to the following documentation:

1. layerNormalizationLayer: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.layernormalizationlayer.html
2. batchNormalizationLayer: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.batchnormalizationlayer.html

For more information on the TensorFlow-Keras layers supported in MATLAB using 'importTensorFlowLayers', refer to the following documentation:

https://www.mathworks.com/help/deeplearning/ref/importtensorflowlayers.html#mw_8c774782-3a98-4dfc-838a-50eb34865d97

Answer 2

[bob scott]

Hi, I would like to to take keras model file that contains a 'tf.keras.layers.Normalization' layer for normalizing inputs to the model and import it and have it 'work' without having to adjust anything on the model side.

Are there any plans to support 'tf.keras.layers.Normalization' when importing keras models?

[Rickard Andersson]

I total agree on this!!