My MATLAB excursion, followed a break with a week in the White Mountains and another week at Old Orchard Beach. I found it hard to get directly back to editing my Mental Construction (MC) web site.
To ease back into thinking about neural network processing in the brain, I searched the web for neural net (NN) packages—mine dated back to a Hopkins AI class in the early 90s. I started searching for hardware implementations. A bit more than 10 years ago, I worked with the Parallax microcontroller chip and although its architecture offered what I needed, between the implementation details (not a strong suit for me) and a growing problem of a tremor in my hands, I had to give up that hardware avenue. I could not find any other hobbyist hardware entry points—at least, not using the terms I searched by.
I searched for Kohonen NNs. Kohonen are neural networks that self-organize which are, imo, at the root of all our thinking. Eventually, I stumbled upon MATLAB, which I had second-hand knowledge from my use of MathCAD in early 2000s. MATLAB offered a neural network package that would allow me to customize the NN parameters, like layer size and learning rate.
Howard County Library
On the Howard County library site, I found they had a link to an online course on MATLAB. I watched the Lynda course for the general approach of the software and discovered it a good bit like python, which I’ve been using for several years. By the way, all Lynda courses (tech training and more) are free of charge to Howard County library patrons through their license.
After reading some on the MathWorks website about their neural network package, I decided to buy MATLAB and a few add-on packages. MATLAB is a programming language to assist engineers and scientists in their handling of large quantities of data. I had to learn that syntax before I could use their Neural Network package.
MathWorks offers an OnRamp training. The first course I took was very similar to the Lynda course, but now I wanted to learn the technique rather than just appreciate its actions. That approach took me five hours, then more hours in Deep Learning. And more to come.
Neural Networks in MATLAB
The neural network package is easy and intuitive to access in MATLAB. Reading through their documentation (I’ve just started this), I discovered some terminology from the Deep Learning and Artificial Intelligence that will help me in MC descriptions. In particular, the distinction used between data classification and data clustering.
- Data classification implies the raw data will be assigned to a known set of categories.
- Data clustering implies the raw data will be gathered into categories that the neural network will generate.
My main interest in Mental Construction (MC) is data clustering, the spontaneous organization of information. I certainly don’t discount data classification, an essential component to handling knowledge we acquire as part of a culture. However to the MC site’s detriment, I drew the distinction between clustering and categorization with personal terminology. I need to replace that with the crisper terms common to the field.
Data clustering, the organizing of data spontaneously, defines the essence of a person’s unique understanding of reality. Data categorization is the method of learning societal mores and standards and most academic learning. Both are essential, but clustering leads to personal, creative perspectives while categorization teaches prearranged relationships.
MATLAB handles tables and inputs spreadsheets so well that my portfolio application (written in python) and the household electric use history (Excel) will probably reveal additional secrets under MATLAB investigation.
My MATLAB excursion has just begun. I expect many fascinating hours to come.
Layers of data clustering (self-organization) and data categorization (cultural) underlie a person’s view of the world.