vslab/DRMH12

DRM-H12

Prompt

-

Model description

DRM-H12-v2 (fix15) — Hierarchical Self‑Model of Consciousness

> Real‑time simulation of 1,156,000 neurons, 27.7M synapses, hormonal dynamics, and integrated information theory (IIT).

This project implements a biologically inspired computational model of consciousness, integrating multiple neurocognitive theories into a single running system.
It is not a classical ML model but a dynamical neural simulation with synaptic plasticity (STDP, Oja, BCM), structural plasticity, a hormonal system with circadian rhythms, and a hierarchical self‑model (HSM) with predictive processing.

The simulation is written in Python (FastAPI + NumPy + Numba) and visualized via a Next.js / React front‑end** with real‑time charts and control panels.

---

🧠 Theoretical Foundations

The model combines five major theories:

Theory	Implementation in DRM-H12-v2
Integrated Information Theory (IIT) – Tononi	Φ(t) computed from network partitions, influences consciousness index Θ.
Global Workspace Theory – Baars & Dehaene	G(t) = tanh(α · mean(x)) – a broadcasted neural activity pool.
Hope Theory – Snyder	H(t) = A(t) · P(t) (agency × pathways) – a meta‑motivational signal.
BCM & Oja & STDP	Tri‑factor synaptic plasticity: Hebbian, homeostatic, and spike‑timing dependent.
Predictive processing + Hierarchical Self‑Model (HSM)	Two‑level self (bodily S1, abstract S2) with prediction errors and reflection depth D(t).

---

📊 Key Model Features

Neural layer

• 1,156,000 neurons placed on a ring topology (RING_HALF = 5000)
• 24 synapses per neuron (sparse, float32) → ~27.7 million connections
• Spiking activity with adaptive thresholds

Plasticity

• STDP (spike‑timing dependent plasticity)
• Oja’s rule (weight normalisation)
• BCM (Bienenstock‑Cooper‑Munro) with sliding threshold
• Structural plasticity – synapse growth and pruning (up to 13%/day)

Hormonal system (10 hormones)

• DA (dopamine), 5‑HT (serotonin), NE (norepinephrine), MLT (melatonin), CORT (cortisol), OXT (oxytocin), EPI (epinephrine), LEP (leptin), T (testosterone), GHR (growth hormone)
• 5 antagonistic pairs with product balance (PAIR_MIDPOINT=0.40)
• Circadian rhythms and anti‑extremal mechanisms

Consciousness metrics

• Φ – Integrated information (Tononi IIT)
• Θ – Full consciousness index = Ω · ξ · Ψ · E_info
  (hormonal duality × motivational conflict × spectral integrity × information energy)
• E_info – Information energy ("fuel of consciousness")
• Self‑referent neuron: S(t) = σ(γ·x + δ·S(t-1))

Sleep/wake cycle

• 1000 steps / day, 15% sleep (150 steps)
• Memory consolidation (weak synapses reset, strong protected)
• Neurogenesis boost during sleep (×1.8)

---

📁 Dataset Acknowledgment – Human Data Used

This simulation was calibrated using real human neurobehavioral data.

We used the Probability Decision‑making Task with ambiguity dataset from OpenNeuro:

> ds004917 – Valdebenito‑Oyarzo et al. (2024)
> PLOS Biology, doi:10.1371/journal.pbio.3002452

The empirical EEG and behavioural responses under ambiguity informed:

• the tuning of uncertainty‑related neuromodulators (norepinephrine, dopamine),
• the structural plasticity rates,
• and the parameters of the predictive self‑model (fix15).

Please cite both the model and the original dataset when reusing or extending this work.

---

🚀 How to Run This Space

This Space uses Docker to run the full stack (Python backend + static frontend build).
When you open the Space, the simulation will automatically start at the exposed port `7860`.

Manual run (local):
If you clone the repository, you can also run it locally:

```bash

Backend

cd backend python -m venv venv source venv/bin/activate # or venv\Scripts\activate on Windows pip install -r requirements.txt python main.py

Frontend (in another terminal)

npm install npm run dev

Download model

Download them in the Files & versions tab.