Jean-Paul CIPRIA


Artificial Intelligence 06 – Neurones Concepts Recognition by 256 Synapses

Written By: Jean-Paul Cipria - Juil• 29•17
Intelligence Artificielle 7 Segments - ©Jean-Paul Cipria 2017

Intelligence Artificielle 7 Segments – ©Jean-Paul Cipria 2017.

How to simulate a pattern recognition ? For exemple we have to learn by training to recognize the 10 first arabian letters to assignate to « zero » to « nine » concepts.
Where are Synapses ? Where are the Neurones ? Where are the Concepts Localization on Brain ?


Created :2017-07-29 17:54:42. – Modified : 2019-06-21 11:56:41.

Difficult ! Master Level.

Difficult ! Master Level.

Done ... Un Café ?

Done … Un Café ?

Let’s see a Matlab simulation for a prototype without using tools elaborate for ?

Video Demonstration in French

I am sorry the video presentation is in French. It’s so difficult to comment it in live directly. Then I explain all other thinks in English texts below.

Intelligence Artificielle 7 Segments - ©Jean-Paul Cipria 2017

Intelligence Artificielle 7 Segments – ©Jean-Paul Cipria 2017


Brain Structures and Functions as described in 1958 [ROSENBLATT-1958] scientific references got us some ideas to represent some « Concepts Zones », some « Neurones » and some « Synapses ». We define 10 concepts : Zero, One … Nine. Those concepts are illuminated by a 7 segments display. Training use to recognize plus or minus one of this 10 concepts and link by a synapse to a neurone a 7-Segment figure to a zone of concept. We have not implemented « slippage analogies » or slip form « 10 concept » to « decimal calculation » from example as inspired by 1995 [HOFSTADTER-MITCHELL-1995].

Définition Intelligence Artificielle 7 Segments - G - ©Jean-Paul Cipria 2017

Définition Intelligence Artificielle 7 Segments – G – ©Jean-Paul Cipria 2017

Like in brain we define memory by 10 metric space local zones in 3D. If we recognize the 7-segments figure ‘2’ with a small imperfection like missing or adding a segment then we « approch » this figure to the concept ‘2’. We arrange all various recognitions around the concept with a small distance in the « Z » distance.

256 Synapses Constructions and Recognitions

In program there is not so AI and AII artificial inputs for dispaching physical information. Your eyes do that with the following animation. It is an illusion. Therefore those inputs are directly sended to th right neurone via a « discriminator » as the role of AII intelligence core. Then we form directly connection to a « fuzzy cabled part » or a « devoted fuzzy area » to store in a metric N Dimension an particular information.

Synapses Constructions - Intelligence Artificielle 7 Segments - ©Jean-Paul Cipria 2017

Synapses Constructions – Intelligence Artificielle 7 Segments – ©Jean-Paul Cipria 2017

Hamming Distance Recognition

The ‘error’ recognition is a one bit Hamming distance. If a figure is not the same when you add or delete a segment then recognition is affected by a 0.9 weight. If a picture is totaly recognize the weight is 1 (100%).

Intelligence Artificielle 7 Segments - Hamming Distance - ©Jean-Paul Cipria 2017

Intelligence Artificielle 7 Segments – Hamming Distance – ©Jean-Paul Cipria 2017

Remarks : A segment is not as a bit because, for example, we can consider the information segment is coded in 7 positions then segment entropy information is about 3 bits. If we think it is in a two spaces there we add a supplementary bit. Therefore in program we consider a segment as a single element of information and we simplify notation.

Synapses Inhibition ?

When a figure is 100% weight, ie totaly recognize, there are some synapses with weight=0.9 to link sensors to neurones. Then we have to construct an inhibition Synapse Algorithm to sleep down and delete those synapses and bring up the correct unique synapse to the minimum distance. After that the « concept distance » from the « two » concept to the ‘2’ picture remains minimal and impulsion for decision remains rapid. CQFD.

Program Construction

It is a challenge to perform artificial intelligence without Matlab tools ? Yes it is possible if we understand correctly what we do. Do we use determinic mathematical link to do this ? Yes and … not. Why ? Because in fact we lay the « logical » or rules as « prolog » can do. That is we have to find some algorithms to find bad and right solutions after passing through rules. Then there are not parallelism nor so much « codelets » [HOFSTADTER-MITCHELL-1995] to perform our 256 possibilities. We just choose to run randomly them and to stop after 10, 20 100 tries. There we have a small « Coderack » as somebody can memorize and visualize in subconscient some inputs with eyes. Then we construct as a « prolog » heart ! Is it a good exercice for scientist ? « Who can do more can do less !« 

How to Construc Parallel Codelets ?

Matlab Begin

% Matlab 4 Cores Processor Parallel Calculations

matlabpool open local;

% Replace the for instruction by a parfor to send part of calculation on one of four processor.
parfor n=1:1:100

End of Matlab

Thus we have a « coderack » with four ideas in our head proceeding simultaneously. It is as great as some person we know ? 😉




Absolutely to read. Even if it is difficult you can understand one or two concepts and doing mathematics descriptions as low level course done it.

  • [ROSENBLATT-1958] : ROSENBLATT, F. « The perceptron: a probabilistic model for information storage and organization in the brain ». Psychological Review Vol. 65, No. 6, 19S8, 1958, 23.
  • [HOFSTADTER-MITCHELL-1995] : HOFSTADTER, Douglas R., et Melanie MITCHELL. The copycat project: A model of mental fluidity and analogy-making. The Fluid Analogies Research group, Fluid Concepts and Creative Analogies. Vol. Basic Books., 1995.


Jean-Paul Cipria.

You can follow any responses to this entry through the RSS 2.0 feed. You can skip to the end and leave a response. Pinging is currently not allowed.

Laisser un commentaire

Votre adresse de messagerie ne sera pas publiée. Les champs obligatoires sont indiqués avec *

Anti-Spam Quiz: