A Julia artificial neural network library which features high level abstraction as well as high performance.
The goal of this project is to create higher order abstractions on the base of ANN.
When come to higher order abstractions, the key concept is rule. For example, we use mathematica rules to carry out our computations, follow the programming rules to write programs, and use grammar rules to organize our thoughts into languages. And the core concept of the rule is axioms. And what are rules about? The rules are about functions of the objects. So we can see that the nature of abstractions is: objects, functions, and rules. Coming back to the mathematica example, numbers are objects, plus and minus are functions, and rules are about how to carry out the functions on the objects.
And then we can come back to our ANN, to use ANN to represent higher order abstractions, we could use layers of artificial neurons to represent objects, matrices and transformations to represent functions, and we can use back-propagation to train our rules. This is the idea of the project.
To run this program, you have to have Julia 0.5. No other dependencies are required.
If you have any questions about the project, don’t hesitate and email to lch34677@gmail.com.