That’s all – support for a neural network

This is my last technical post in the Get Noticed 2017 category. The last one will be a quick summary of the last three months. Soo…it was a very long week. I didn’t do too much but there was a big progress in using the neural network in the Aksesi Proxy Application.

I would like to remind you that all of the sources are available in my Github repository. Moreover, in the README file, you can find short project description with its main assumptions.


I created a simple class which is responsible for preparing the NN at the application’s startup. The whole magic happens in one method – trainNeuralNetwork

As you can see, the trainingService is used. In that class, I encapsulated things related to Encog training process:

Unfortunately, doing this training at the startup extended application’s bootstrap to almost 5 minutes.

The neural network in the application is represented by a component which contains the network’s object:


The integration process was fairly simple. Everything I had to do was to implement the  IConversionStrategy interface. The convert  method has the following body:

There are a few well-known bugs/issues. As you can see, it returns wrong value, but for the development purposes, I used logs.


The neural network integration architecture is prepared. Unfortunately, the network does not work. It returns wrong results. I guess it is connected with its configuration (a number of layers, a number of input values for each of layers, activation functions). It would be nice to prepare infrastructure for researching the best configuration with the lowest error factor.


As I mentioned at the beginning, it is the last technical post in the Get Noticed category. The Aksesi application is completed in 98%. The frontend module works without any issues. The backend service is able to resolve basic gestures and communicate with authentication endpoints. Moreover, it uses the neural network for gestures recognition.


You may also like