Static Neural Network

What is this?

It's a strongly typed neural network. Models' architectures are decided at compile time. If this seems like an unnecessary hassle, it's because it is. Tell me if you find a use case!

What does it do?

You can assemble networks by layering dense or convolutional layers with a selection of activation functions. You can optimize them with a selection of gradient descent backpropagation algorithms, with respect to a selection of loss functions.

How do I use it?

Check examples/mnist/

Dependencies

• CMake 3.10 or later
• C++17-compliant compiler

Setup

$ mkdir build
$ cd build
$ cmake ..

Build and run

Use your favourite build system. This is how I do it:

Visual Studio (Windows)

• Open the solution file in build/
• In the Solution Explorer, right click e.g. MnistExample and Set as startup project
• Click Local Windows Debugger to build and run.

Make (Linux)

(Starting from the build/ directory)

$ make
$ (cd examples/mnist && ./MnistExample)