This is a simple Ternary Neural Network to demonstrate how TNNs work, using the 'MNIST' and 'CIFAR10' datasets for simplicity. This will only take a few minutes. To train on MNIST, simply run:
$ python3 /app/Soteria/TNN/main.py
This will save a model_architecture.dat file containing the architecture components list, and weights.dat file containing the ternary weights of the trained model.
Simlarly, to run TNNs over 'CIFAR10' dataset, run:
$ python3 /app/Soteria/TNN/main_cifar.py
You can create different architectures in the respective files to train them.