Repository files navigation

The Quantum ReLU'QReLU' and its modified version or'm-QReLU' are Python custom activation functions available for both shallow and deep neural networks in TensorFlow, Keras, and PyTorch for Machine Learning- and Deep Learning-based classification. They are distributed under theCC BY 4.0 license.

Details on these functions, implementations, and validations against gold standard activation functions for both shallow and deep neural networks are available at the papers:Parisi, L., 2020 andParisi, L., et al., 2022.

The dependencies are included in theenvironment.yml file.Run the following command to install the required version of Python (v3.9.16) and all dependencies in a conda virtualenvironment (replace<env_name> with your environment name):

• conda create --name <env_name> --file environment.yml

Run pip install -e . to install thesrc package in editable mode.

You can use theQuantumReLU activation functions as a keras layer and set themodified attribute to eitherFalseorTrue if using the QReLU or the m-QReLU respectively:

Either

model=models.Sequential()model.add(layers.Conv2D(32, (3,3),input_shape=(32,32,3)))model.add(QuantumReLU(modified=False))# True if using the m-QReLU (instead of the QReLU)model.add(layers.MaxPooling2D((2,2)))

or

model=keras.Sequential(keras.Input(shape=(32,32,3)),layers.Conv2D(32,kernel_size=(3,3)),QuantumReLU(modified=False),# True if using the m-QReLU (instead of the QReLU)layers.MaxPooling2D(pool_size=(2,2)),    ])

self.conv1=nn.Conv2d(1,32,kernel_size=3)self.relu1=QuantumReLU(modified=False)# True if using the m-QReLU (instead of the QReLU)self.pool1=nn.MaxPool2d(kernel_size=2)

isort is used to ensure a consistent order of imports, whilstautopep8 to ensure adherence of the codes to PEP-8,via the following two commands respectively:

• isort <folder_name>
• autopep8 --in-place --recursive .

Runpytest --cov-report term-missing --cov=src tests/ to execute all unit tests and view the report with the testcoverage in percentage and missing lines too.

If you use these activation functions, please cite the papers byParisi, L., 2020 andParisi, L., et al., 2022.