------------------------------------------------
# english
------------------------------------------------

UI & Execution Instructions
To run the executable, please unpack the .zip file and ensure all satellite files are saved in the same directory as the executable (.exe).

Image Overview
Structure: The image displays the circular architecture of the Fibonacci layout.

Fovea: Located at the center (marked in green).

Prime Numbers: Hollow yellow dots represent Fibonacci nodes that correspond to prime numbers.

TRN Inhibition: Red-dotted zones correlate with attention mechanisms—specifically, regions inhibited by the Thalamic Reticular Nucleus (TRN).

Functional Breakdown
A. Biological Architecture
The system simulates a complex neuromal architecture modeled after the human brain. Typically, visual and auditory inputs require multi-level correlations. In human biology, the right hemisphere receives vertical hemi-field inputs from the right half of both eyes, while the left hemisphere processes the opposite halves. These hemispheric patterns are synchronized via the corpus callosum. Since this demo utilizes a single camera input, these complex binocular visual integration operations are bypassed.

However, the demo effectively showcases the cortico-thalamic loop. The camera input is decomposed by color and mapped onto a pre-calculated Fibonacci (Golden Ratio) matrix. The central, high-density area mimics the human fovea (which covers roughly 1% of the visual field). This zone processes Red and Green channels via dense satellite nodes indexed by blue anchor nodes.

The signal path travels from subcortical structures (Thalamus/LGN) to the Cortex, where it is processed and retransmitted back to the thalamus through the TRN. The incoming video stream is continuously compared and partially inhibited; this cortical inhibition forms the structural basis for Bayesian prediction. Blue light channels (the Koniocellular pathway—Casagrande et al.) are treated with maximum priority to rapidly close this loop, acting as the structural backbone for data addressing and information processing.

B. Computational Engine
The cortico-thalamic loop is implemented as a ring buffer processing a 64-bit vector (analogous to a cortical column). The ring buffer's sequence is driven by Koniocellular indexing, where retinal node addresses are bound to a fixed coordinate system tied to the blue nodes.

Operating as an asynchronous system with parallel-activated processing zones, the engine computes camera signals in float32 without accumulating rounding errors. The architecture strictly avoids both Transformer-based models and backpropagation. The demo code is heavily streamlined; the production version runs without visual window rendering (as rendering the retina layout would introduce unnecessary processing latency) and relies entirely on Shannon entropy computations across cortical structures.

C. LLM Integration
If you are unfamiliar with the underlying neurodynamics, you can upload the provided Python files (neuron_engine.py and brain_schema.py) along with this documentation or the demo source code into your preferred LLM. All tested Large Language Models are capable of delivering granular explanations of the b4 architecture when prompted with these files.

Note: The author is a physicist—neither a neurologist nor a professional programmer. Consequently, there is immense room for improvement and optimization (alongside upgrading to better hardware).

------------------------------------------------
# romanian
------------------------------------------------

Pentru a rula executabilul, va rugam sa despachetati fisierul .zip si sa salvati fisierele satelit in acelasi director cu fisierul executabil (exe).

Ce vedeti pe imagine:

In primul rand vedeti forma circulara a structurii Fibonacci. In centru (cu verde) se afla foveea. Cu puncte galbene goale sunt marcate punctele
din structura Fibonacci care corespund unor numere prime. Zonele cu puncte rosii sunt corelate cu atentia, adica zonele inhibate de TRN. 


Cateva explicatii functionale:

  a. Biologic.
  Sistemul simuleaza o structura complexa, similara creierului uman. In mod normal imaginea/sunetul de la ochi/urechi, necesita corelatii multiple.
Pe emisfera dreapta ajunge semnalul de la jumatate (pe verticala) din ochiul drept si jumatatea dreapta de la ochiul stang. Pe ochiul stanga se coreleaza
imaginile de la celelalte jumatati de ochi (hemicampuri). Imaginile din emisfere se sincronizeaza printr-o structura de conexiuni speciala numita 
corpul calos. In cazul rularii demo se foloseste un singur 'ochi' (o camera) asadar anumite operatiile complexe de integrare vizuala nu au cum sa se intample.
  Cu toate acestea, rularea demonstrativa nu este superflua, ea se foloseste de un alt loop, numit bucla cortico-talamica. Pentru inceput, imaginea de la camera
este descompusa pe culori si mapata pe o structura calculata la inceputul simularii, de tip Fibonacci (sectiunea de aur). Pe ochi, exista o zona centrala densa
numita (la om) foveea, care acopera o suprafata infima din campul vizual (cam 1%). Din aceasta zona se citesc si culorile Rosu si Verde (puncte satelit dense, situate
si indexate dupa 'nodurile' albastre.
  In creier, semnalul parcurge prima data structurile subcorticale (talamus -la noi LGN- si altele) apoi ajunge in cortex, unde este procesat si retransmis catre talamus prin intermediul unei structuri specializate numita TRN. Imaginea venita de la camere este in permanenta comparata si partial inhibata, inhibitia corticala formand
asa-numita predictie bayesiana. Lumina albastra (calea Konio- a nisipului- Casagrande & al) de pe camere este procesata cu prioritate maxima, in asa fel incat sa inchida rapid bucla mentionata, acesta constituind scheletul structural de adresare si procesare a informatiei.

  b. Computational
  Bucla cortico-talamica este un ring buffer care proceseaza un vector de dimensiune 64 de biti (similar cu o coloana corticala). Ordinea din ring buffer este generata
prin indexare konio (adresele nodurilor de pe retina sunt indexate dupa un sistem fix de referinta legat de adresele nodurilor albastre). Sistemul poate procesa semnalele
de la camera in float32 fara sa acumuleze erori de rotunjire datorita faptului ca sistemul este asincron, adica activeaza in paralel diverse zone de procesare. Sistemul
nu foloseste un model de tip transformer si nici backpropagation. Codul pentru demo este simplificat. Aplicatia de lucru functioneaza fara vizualizari (randarea ferestrelor
de tipul retinei ar intarzia nejustificat procesare), doar cu calcule de entropii Shannon pe structurile corticale. 

  c. LLM
  Daca nu esti familiarizat cu domeniul, poti sa incarci in fereastra LLM-ului tau preferat fisierele cu extensia python (.py) numite neuron_engine si brain_schema (incluse
in fisierul zip) impreuna cu documentul de fata sau codul sursa al programului demo. Toate aplicatiile de tip LLM testate pot furniza explicatii detaliate despre felul in care
functioneaza b4 pornind de la aceste explicatii.   
 
Nota: Autorul este fizician, nici neurolog, nici programator. Asadar exista mult loc pentru imbunatatiri (pe langa hardware mai bun).