Binary Neural Networks

• Sample project used for coljure meeting.
• The slides can be found from Google presentation
• All networks use supervised learning strategy.
• This project trains neural networks for or, and and xor gates.

Prerequisites

• [Leiningen][1] 2.0 or above installed. [1]: https://github.com/technomancy/leiningen
• Clojure 1.8

Running

To start a web server for the application, run:

lein run

Open your browser and access to: http://localhost:3000/swagger-ui/index.html

Considerations

• The learning-factor for all gates is 0.5
• The Bias is conformed by X0=-1 and umbral (theta) value. It starts in 0.4 for each neural network.
• The weight's correction is done by:

  wi = wi + 2 * learning-factor * yi_expected * xi;

• The umbral value is modified by:

theta = theta + 2 * learning-factor * yi_expected * X0;

• The active function uses tanh.
• The stop criterion is 100 iterations.

License

Copyright © 2017 Apache