Skip to content

Conceptual diagram of whole brain reproduction system — EvoSpikeNet complete document

[!NOTE] For the latest implementation status, please refer to Functional Implementation Status (Remaining Functionality).

Creation date: 2026-02-26 Last updated: 2026-02-26 Version: v0.4.0 (Feature 13 Fully Integrated + Remaining Feature Details) Author: Masahiro Aoki (Moonlight Technologies Inc.) Target audience: Brain scientists, neuroscience researchers, system architects, AI researchers

This document is a "system that reproduces every human brain" based on EvoSpikeNet. Comprehensive presentation of concepts, architecture, data flows, and mathematical models. In addition to the implemented functions (v0.3.x), the functions listed in Remaining_Functionality.md Future functions will also be integrated. Specify the neuroscientific basis and paper citations for each section.

1. Overall overview and purpose

The human brain consists of approximately 86 billion neurons and approximately 100 trillion synapses connecting them. It is a huge parallel computer. EvoSpikeNet currently has several areas and functional modules. Implemented as a distributed node, in the future "whole brain reproduction" covering all areas Aim for.

  • Goal: Maintained some degree of biological plausibility while maintaining specialization by brain region. Build a spiking neural network group and a distributed control layer that controls it.
  • Scope: Sensor input → recognition → decision making → motor output + long-term memory/evolution/self-repair Including cycles. In future phases, we plan to add the hippocampus, amygdala, and cerebellum.

Quote: - Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Transactions on Neural Networks. - Maass, W. (1997). Networks of spiking neurons: the third generation of neural network models. Neural Networks. - Buzsáki, G. (2005). Theta rhythm of navigation: link between path integration and landmark navigation, episodic and semantic memory. Hippocampus.

2. Conceptual diagram/architecture

The Mermaid diagram below shows the logical configuration of the "Whole Brain Reproduction System". Implemented modules are in bold, To be added in the future (see Remaining_Functionality) is written in italics.

graph LR
    subgraph "Sensory Layer"
        CAM[Camera] -->|visual signal| VIS("Visual Module")
        MIC[Microphone] -->|auditory signal| AUD("Auditory Module")
        PROPRIO[Proprioception (planned)]
        OLFACT["Olfactory Module (planned)"]
    end

    subgraph "Encoding Layer"
        VIS --> TAS(TAS-Encoding)
        AUD --> TAS
        OLFACT --> TAS
    end

    subgraph "Cognitive Layer"
        TAS --> SLM["Spiking LM\n(language/multi-modal)"]
        SLM --> RAG["Hybrid RAG"]
        RAG --> PFC["Q-PFC control"]
        PFC --> MTR["Motor Planner"]
        PFC --> AMY["Amygdala (emotion) (planned)"]
    end

    subgraph "Memory/State"
        EPI["Episodic Memory"]
        SEM["Semantic Memory"]
        MINT["Memory Integrator"]
        HIPP["Hippocampus (planned)"]
    end

    subgraph "Evolution & Recovery"
        EVO["EvoGenome Manager (planned)"]
        REC["Auto Recovery Engine (planned)"]
    end

    subgraph "Infrastructure"
        ZEN[Zenoh Pub/Sub]
        PTP[PTP Sync]
        GEO[GeoNodeManager]
        AUDT[Audit/Integrity Logging]
    end

    TAS --> SLM
    SLM --> RAG
    RAG --> PFC
    PFC --> MTR
    PFC <--> EPI
    PFC <--> SEM
    EPI <--> MINT
    SEM <--> MINT
    MINT --> RAG
    ZEN <-- PFC
    ZEN <-- VIS
    ZEN <-- AUD
    ZEN <-- OLFACT
    ZEN <-- AMY
    ZEN --- PTP
    ZEN --> GEO
    ZEN --> AUDT
    GEO --> "Global Routing"

2.1 Overall data flow

  1. Input: Sensor data is converted into a temporal spike train by TAS encoding ([Izhikevich03]).
  2. Cognition: Spiking LM promotes cross-modal attention (ChronoSpikeAttention; [unpublished paper]) integrated using. The formula will be explained later.
  3. Search: Hybrid RAG retrieves context from memory.
  4. Control: PFC calculates Q-PFC self-gate.
  5. Output: Motor Node generates a motion command.
  6. Memory: Episodic/semantic storage of experiences.
  7. Learning: Synapse update with STDP/Meta-STDP
  8. Evolution: Structural adaptation (Remaining Feature) by EvoGenome.
  9. Self-healing: AutoRecovery resets parameters and relocates nodes on failure.

3. Mathematical models and formulas

3.1 Neuron model

a) LIF $\(\tau_m \frac{dV}{dt} = -(V-V_{rest}) + RI(t), \quad V\ge V_{th} \Rightarrow spike.\)$ (Izhikevich03)

b) Izhikevich $\(\begin{aligned} \frac{dv}{dt}&=0.04v^2+5v+140-u+I,\\ \frac{du}{dt}&=a(bv-u),\\ if\,v\ge30&:\;v\leftarrow c,\;u\leftarrow u+d. \end{aligned}\)$

c) ChronoSpikeAttention (patent MT25-EV001): $\(Attention(Q,K,V)=\mathrm{softmax}\left(\frac{QK^T}{\sqrt{d_k}}+\alpha e^{-\frac{(t_i-t_j)^2}{\sigma^2}}\right)V.\)$

3.2 Spike encoding

TAS-Encoding ([Unique to this project]): $\(\mathrm{TAS}(x_t)=\sum_{k=1}^N \delta(t-t_0-\Delta k)[x_t=k].\)$

3.3 Evolutionary DNA representation

EvoGenome converts the gene sequence \(G\) $\(G=\{w_{ij},\theta_k, \tau_l\}\)$ The fitness \(F(G)\) is evaluated for each generation and mutation/crossover is performed. (Details: evospikenet/evolution.py)

3.4 Distributed communication model

Zenoh Pub/Sub measures the latency \(L_{ij}\) between nodes \(i,j\), The goal is \(L_{ij}<50\mathrm{ms}\). GeoNodeManager is a region Solve the optimal routing using the delay matrix \(D_{rs}\) between \(r\).

4. Essay/material citation

  • Izhikevich, E. M. (2003). Simple model of spiking neurons. IEEE Trans. NN.
  • Maass, W. (1997). Networks of spiking neurons .... Neural Networks.
  • Buzsáki, G. (2005). Theta rhythm of navigation ... Hippocampus.
  • Pfeifer, R., Iida, F. (2004). Embodied Artificial Intelligence. Trends in Cog Sci.
  • O’Reilly, R. C., Munakata, Y. (2000). Computational Explorations in Cognitive Neuroscience.
  • Internal documents such as Atsugihara papers (EvoSpikeNet patent application materials).

5. Future features to include from Remaining_Functionality

  • Spatial perception/generation system (Feature 13): Spatial Where/What/Integration/Attention nodes.
  • Geographically distributed management (Feature 40): Multi-DC deployment and latency matrix.
  • Automatic Recovery (Feature 36): AI-based anomaly detection and recovery playbook.
  • Audit Log / Tampering Detection: Activity record using SHA-256 hash chain.
  • Evolutionary Genome UI/API/SDK: Users can directly manipulate the DNY structure and perform adaptive experiments.
  • High precision space generation: Improved spatial module and 4 dedicated nodes (Rank12-15) 3D spatial mapping.

By integrating these, the conceptual diagram of this document is ``Human Whole Brain Simulation''. Runs on general-purpose clusters with geographic redundancy, self-healing, and evolutionary adaptation "Software Platform".

6. Future prospects

  • All areas covered: Additional implementation of cerebellum, amygdala, olfactory cortex, etc.
  • Biopotential Interface: Real-time closed-loop control with EEG/BMI.
  • Federated learning: Distributed learning across multiple facilities and countries.
  • Hardware implementation: High-speed spike calculation using FPGA/ASIC.

This document is a conceptual document for research and design; the implementation code is

evospikenet/ See below.