Is "fusing batch normalization and convolution in runtime" supported in deep learning?

Currently pytorch deep learning framework is supported, is matlab supported in the future?

Answer 1

[Shreshth]

Hello,

Currently, MATLAB’s Deep Learning Toolbox does not support fusing batch normalization and convolution layers at runtime directly. This technique is typically used as an inference-time optimization to improve runtime performance by merging the batch normalization parameters into the convolutional layer’s weights and biases1.

However, MATLAB does support batch normalization layers, which can be used between convolutional layers and activation functions to speed up training and improve network performance.

Please refer this link regarding batch normalization layer for more information.

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.batchnormalizationlayer.html

I hope this helps!