You are here

Building and visualising the Neural Networks

Naam bedrijf: 
ToThePoint

Building and visualising the Neural Networks
Keywords
• Machine Learning (Neural Net)
• Java (our neural net is written in java, but anything that can work together with our NN would work)
• Literature review
• API review
• Not afraid of some basic mathematics

Level
Mid-high

Prerequisites
The student(s) should, preferably, have a basic knowledge of machine learning and java programming, UI building.

Description
This internship project consists of building a neural net, GUI to intervene on its parameters and visualising it on screen while making decisions.

Currently we are building a neural network - POC for our home-made arcade play console, which will enable us to recognise the game a player is currently playing on the arcade machine. There are various machine learning libraries out there, but we start building it ourselves just for the sake of it.

Our ultimate goal is to visualise the Neural Network so that its parameters can be adjusted using a GUI. Number of hidden layers, number of neurons, learning rate, parameters for convolution layers etc. Also as soon as the Neural network is built we should be able to show it’s decision process on screen in realtime.

What would would be nice to visualise?
• How its convolution layers (if any) are built up
• Decision boundaries
• Neural Network (layers, neurons, weights)

As previously stated, no API’s are involved, the Neural Network is home grown so can easily be extended.
Our approach of this internship project will be iterative, first start the student will start with the simplest possible perceptron to solve the XOR problem and then we can start adding complexity to it.

The deliverable would be an approach or a start of how we can extend it to our arcade machine so that we can experiment with our Neural Network to find the best setup to train our neural network.
And if possible show in realtime how the Neural Network is built up while playing the game and recognise the game while it is being played.

For the visualisation part, the student is free to visualise the network how he wants, requirements are that it is easy to grasp for presentation purposes to none-specialised people. There are various API's around that the student is free to study and use.

Sponsoring
This internship project is sponsored by the ToThePoint subgroup of the Cronos group. We expect the student not to be afraid of tackling a challenging programming task.

Soort betrekking: 
Departementen Erasmushogeschool Brussel: