Since the beginning of the Arduneuron Series there have been many breakthroughs in my understanding of how neural networks work and how to implement them better into my projects. I would change slight things in the theory of the neural network. Most of my ideas were programmed and never even compiled or put into anything else, they were just stepping stones in my understanding.

Eventually I came up with a method that allowed me to implement any learning style that I wanted and make any architecture for neural networks that would work for a certain situation. I finally made the tenth generation of Arduneuron, the generation that I changed the name to Mechaneuron.

The reason for the change in the name was because I was working on implementing the neural network into an operating system that would use these neurons. I tried to make a series of libraries that would work in the Arduino environment and allow me to make physical robots using open-source devices.

All this, of course, was programmed by my hands. No one else ever even tried to help me code my projects because they felt that they could not come even close to how fast and efficient they said I was at programming. That was the most frustrating part: nobody was willing to help. 

Everything in this student organization has been that way. Everyone is out to do there own thing, not even to help on group projects. I couldn't get anyone to work on anything and that's why I decided to go dormant in my responsibilities with ODU Robotics. I still get people coming up to me that say that they've been trying to join up. They said that they've asked around and heard myths about us and our projects, I say us but it was really my projects.

I got people to work on my projects because no one would come up with practical and doable projects that required group work. I had to come up with ideas for an entire year for us to do. It wasn't "us", it was purely "me". Now the group has turned into nothing but a bunch of rusting, robotic skeletons.

Mechaneuron (ArduneuronG10) has had some improvements made to allow the construction of Feed-Forward Neural Networks as well as the integration into C++ image libraries such as CImg. The push to make me design this was discovered after I had a conversation with a guy in the coffee shop I often go to, Borjo Coffeehouse. He has a side hobby of buying and selling tickets to events. I told him that I could develop a software that could bypass the CAPTCHA's and be able to buy multiple tickets faster than a human operator.

My overall goal was to be able to develop an application for Mechaneuron involving images. If I get this working, I'll be using it in reading images and all sorts of other things, including further implementation for reading signs on the road and the like. I'll be taking a hack at it and seeing what I can do.

I have just now completed the alpha stage of a C++ library that I have been working on for the past 2 months. This library enables the user to implement natural selection into a program. The library makes a biosphere with a population of organisms that all have DNA and reproduce sexually. The DNA is transcribed, makes mRNA which is translated and makes Proteins that are determined by the user what their id is and their function within the program. Each organism reads and backs-up its DNA code into a text file.

With the main programmer, myself, being now at my humble abode in the Appalachian Mountains, I have been impaired of my ability to conduct myself into the testing and fixing of bugs into G5. I have only managed as few as about 3 hours of programming over the past week (as compared to over 30 during my summer semester at ODU).

So here the update:
G5 seems to be wanting to stabilize at some point that is dependent on the firing threshold, and doesn't seem to be relevant to the actual input given to the brain. I keep looking through the code and cannot find what could be causing this.
The brain also cannot run if even one neuron has no axons connecting to it, so I have to make it so when has nothing it does nothing.

Whoo, boy! This one definitely took some time thinking over. I even wrote several pages of notes in my college ruled notebook. But once I got the idea down, everything seemed to go as slick as butter. This one is a crossbreed of what I've learned through all the past Arduneurons and added some tricks with the help of Linear Algebra, Differential Equations, and Calculus, oh and Wikipedia as always.

This one was easier to program than any before. It took a week to think about and two hours to program. Holly Cow, that was so much simpler, I don't know why I decided to go with those giant for loops and check every neuron if it pulsed, record it in a checklist, then go to the neurons on that checklist, and blah da blah da blah. Point is, way easier, hopefully it's just as easy to use. I still need to test it.

And another thing, for being called "Arduneuron" three generations didn't allow their incorporation into Arduino, but I found out how to change this and put the "Arduino" back in "Arduneuron".

Partial differential equation implementation, 
Time-driven computing, 
Non-realistic determination of threshold breach, 
Back-propagation using gradients, 
Dynamic learning constant for each neuron,
Dynamic desired output and Mean squared error function,
Floating point value weights, and
Complex feed.

G5 will be a bringing together of lessons learned through G1, G2, and G3.

I have been making a version that is based off of G3. I have this one being a programming language called Thinker, although I don't know if it would actually be applicable and was merely for my first time try at making my own programming language. It is based off of P" but instead of manipulating pointers and the values associated with them, it goes between what neuron in the brain your talking about and adding pulses to them and other things. 

It seems that the G3 and G4 are having a bug though and I can't quite find out where exactly my problems lie.

I will soon implement some new techniques I discovered as I waited for my MAE 205 class to start. This will go into a version based off of G3.

Another thing... I forgot to mention that Arduneuron is only about neural networks and will not have any swarm, genetic, or any other AI techniques about them. I will make different series for each of those, but I want to master the art of brain making before I the other kind. 

Genetic Intelligence has been going through my mind recently and will need to pretty much be a programming language that controls the layout of the brain but at the same time can allow proper gene splicing and all that good mutation stuff.

ArduneuronG3 is currently in alpha stage and is being tested for its ability to multi-thread, neural architecture ease, and new process of delivering data to neurons both between environment and between the neurons themselves.


Hopefully the road to beta will be brief and without many errors that ArduneuronG2 had with it's alpha stage.

ArduneuronG3, the next step of neural network artificial intelligence Arduneuron, will have a new programming language specifically for the neurons and sub-brains of the AI. This will allow easier transfer of data between the neurons and allow complex sections of the brain that are able to communicate to each other. This also allows the user to implement neural architecture into the AI mainframe.

Another new part is the implementation of different types of neurons into the network, allowing the user to construct a complex neural architecture. It seems that as the Arduneuron series progresses more and more input will be needed from the user to make a practical brain to be efficient for the task desired. Some of the neurons being introduced are Constant Spiking Neurons (emit constant spike frequency), Threshold Spiking Neurons (emit one spike once threshold reached), Sub-threshold Excite Post-threshold Inhibit Neurons (what the name suggests), and many more.

The new programming language is being built in hopes of making the neural architecture easier to implement and construct. This means that you have to have some knowledge of what your doing before you can just make a brain. I plan on making both text and video tutorials on the matter once ArduneuronG3 has gone through it's beta phase. ArduneuronG3 right now is only in sub-alpha stage.

Keep tuned for the latest in the Arduneuron series! I may be taking a few days break from this so the next time you hear anything new may be this weekend.