Initial Ideas

The diagram bellow represents an initial idea for a FPGA computational device analogous to a hypercolumn (HC). The FPGA itself would contain the initial algorithms related to pattern recognition and decision making, successful matches and winning decisions data would then be stored on the SRAM to be used by the next HC interaction (or firmware update). The included µcontroller manages the reprogramming via the ISP protocol using the native JTAG interface found in FPGA devices.

Both HC and µcontroller have dedicated communication buses, where the HC addresses other HCs and works in parallel; µcontrollers would receive firmware updates to be programmed via a bus addressable by the HCs. It is important to note that data belonging to each set does not travel via the same channel, with the only exception being when HCs submit a new firmware to one or many µcontrollers.

A block diagram for a FPGA based hypercolumn

Introduction

The purpose of this blog is to document the research of, and possible development of a neuronal network based computational device analogous to hypercolumns using mainly FPGA technology.

The neocortex is the largest part of the human brain (and mammals in general), in humans it is about 3mm thick and with a neuron density of 10⁵ mm^-2, except for the V1 area. The average amount of neurons in the human brain is in the order of 2 x 10¹⁰ and forming a total of 1.5 x 10¹⁴ synapses. There are in the human neocortex about 2 X 10⁶ hypercolumns each composed in average of 100 minicolumns which in turn are made of anything between 80 to 150 neurons.

Hypercolumns (or Cortical Columns) are the building blocks of the mammalian and reptilian neocortex, with a repeatable pattern of structured neurons that are intrinsically related to information gathering, processing and memory; hypercolumns are in essence highly parallel and non-linear computing devices.

As with anything in nature, the human neocortex can be expressed in the form of maths; computational algorithms then are a perfect way to describe and predict the way in which information is processed within the brain.

FPGAs or Field-Programmable Gate Arrays, are very sophisticated devices that can be reconfigured "on the fly" to almost any kind of logic or (to some extent) analogue circuits. FPGAs have advanced immensely in the past few years, with the top models having close to millions of logic gates and embedded microcontrollers, while the price per unit has fallen considerably. FPGAs can be coarsely described as "soft processors".

This project will be initially based on the outstanding work of Christopher Johansson and Anders Lansner from the Department of Numerical Analysis and Computer Science, Royal Institute of Technology of Stockholm, published initially circa 2006.

Personally, I am not involved at the moment with any academic department or institution, and have not any intention to copyright or patent anything I am remotely able to develop. I have been in the computing field for more than twenty years and have a formal background in industrial automation. Electronics and computing have always been part of my life one way or the other, be it work or fun.

I have been recently profoundly inspired by the Singularity Institute to continue and improve my education in my field, and will quite possibly take up to a master degree related to artificial intelligence. In the past I have metaphorically been running aimlessly for an idea that would challenge my capabilities and enable me to contribute to something of real value to society, not just gimmicks powered by consumerism.

There is a very long road ahead of me, in learning and in development, and I am quite certain that I will be incorrect in many aspects of this project, but what really matters is that I will be one way or the other contributing to something bigger than humanity itself.

If you are related to the artificial intelligence field or are just as exited as I am about the possibility of fully conscious artificial minds, feel free to get in contact with me via this blog.