NALU implementation and experiments:
A TensorFlow and Keras implementation of neural arithmetic logic unit from
. The neural arithmetic logic unit is a newly developed logic unit that attempts to improve a neural network’s competence to handle numerical data.
Mixture of Softmaxes implementation and experiments:
A Keras implementation Mixture of Softmaxes. MoS layer is a way training an ensemble method without the having to train the full NN. This layer is described in
Breaking the Softmax Bottleneck