2018-07-31

Artificial Intelligence

My efforts on the fields of artificial intelligence include:

  • Genetic algorithms
  • Neural Networks with supervisor
  • Ant theory
  • Triples theory
  • Bussines Social algorithm

So far, I've made the following implementations on that subjects.

Implementations

Since I hate doing ad-hoc single-purpose implementations, I always try to do as much generalization as possible or reasonable. Java and JavaScript implementations are fully object-oriented. SQL implementation is currently limited to feed-forward due to optimizations, but can be easily tweaked to be generally usable for any type of network. I've implemented most of other AI's paradigms, but those are just proofs of concept.

Neural Networks and Backpropagation in JavaScript

My first implementation of neural networks. Started as a proof of concept, later it became a solid OOP model of neural networks and teaching methods.

Here you can download the files. I didn't write any documentation for the JavaScript implementation, but since the model is the same as in Java, you can see the concept in Javadoc documentation (see and download below).

DOWNLOAD

Neural Networks and Backpropagation in Java

JavaScript implementation re-written to Java. Thus, much faster. Otherwise it keeps the original idea of maximum independence of distinct parts of the model (e.g., computing versus teaching).

What's implemented? Not much, actually:

  • General neural networks stuff like Neurons, Synapses, Transfer functions etc.
  • A perceptron networks generator, which creates perceptrons based on layers definition.
  • A Teacher, or „Supervisor“, which teaches the neural net using backpropagation.

I didn't have time to lead the project to a state when anyone could use it as easily as Joone or OpenAI's imple­mentation, but this can be very appropriate when your needs are simple, as for school projects and simple web applets.

This implementation was made as a part of a school AI project.
It's description can be found here.
It's JavaDoc documentation can be found here.

DOWNLOAD

Anyone is welcome to implement more concepts – like recurent networks, Kohonen maps, Hopfield network etc. The existing model is suitable for this.

Neural Networks and Backpropagation in SQL Stored Procedures

My last work in fields of artifical intelligence is an implementation of neural networks and in SQL. Using a library of Stored Procedures, you can create neural networks, teach them using backpropagation algorithm, and then use them, all just by calling very simple Stored Procedures API.

Further info about SQL implementation of neural networks on a dedicated page.

Other stuff

These are just some humble ideas.

Plan to develop artificial cognition – what has to be done
Covering Artificial Intelligence Theory
Business Society Algorithm


0