This is the Tensorflow implementation of deep semi-random units proposed in [1].
The file semi_random.py contains the semi-random units for convolution and fully connected layers.
The files mnist.py and cifar10*.py are modified from the examples of Tensorflow to demonstrate how to use the semi-random units on MNIST and CIFAR10 datasets.
Reference
[1] Kenji Kawaguchi, Bo Xie, Le Song. Deep Semi-Random Features for Nonlinear Function Approximation. arXiv:1702.08882 https://arxiv.org/abs/1702.08882