“SPOCU”: scaled polynomial constant unit activation function.

Non-official Pytorch/Tensorflow implementation of the SPOCU activation function [1], for the case when c=infinite.

Pytorch

It can be included in your network given an alpha, beta and gamma value:

from spocu_pytorch import SPOCU

alpha = 3.0937
beta = 0.6653
gamma = 4.437

spocu = SPOCU(alpha, beta, gamma)

x = torch.rand((10,10))
print(spocu(x))

Tensorflow

from spocu_tensorflow import SPOCU

alpha = 3.0937
beta = 0.6653
gamma = 4.437

spocu = SPOCU(alpha, beta, gamma)

  
X = tf.Variable(tf.random.normal([10, 10], stddev=5, mean=4) )
print(spocu(X))

Tests

See spocu_test for equivalance of pytorch and tensorflow implementation.

Citation

If you find this work useful, please cite:

@article{carrillo2021deep,
  title={Deep learning to classify ultra-high-energy cosmic rays by means of PMT signals},
  author={Carrillo-Perez, F and Herrera, LJ and Carceller, JM and Guill{\'e}n, A},
  journal={Neural Computing and Applications},
  pages={1--17},
  year={2021},
  publisher={Springer}
}

Acknowledgements

Thanks to the author of the Tensorflow version, Atilla Ozgur.

Bibliography

[1] Kiseľák, J., Lu, Y., Švihra, J. et al. “SPOCU”: scaled polynomial constant unit activation function. Neural Comput & Applic (2020). https://doi.org/10.1007/s00521-020-05182-1