drmylesgarveylabs/ion_neural_network

🔮 Shadows of Consciousness: Ion Neural Networks

The Paradiagm of Ion Neural Networks (INN) -- breaking the voltage abstraction.

Developed by Dr. Myles Douglas Garvey (Independent Researcher) - drmylesgarvey@gmail.com

Currently seeking full-time and part-time employement opportunities.

---

📌 Table of Contents

• 🧠 Overview
• 🏗️ Architecture
• Technical Specifications
• 🧪 Experimental Matrix
• 📊 Evaluation & Benchmarks
• 💾 Dataset & Training Specs
• Hyperparameters
• ⚠️ Limitations & Disclaimers
• 📜 Citation

---

🧠 Overview

NIRGEN (Neurotransmitter Ion Receptor Glial Endocannabinoid Network) is a discrete, biophysically grounded Ionic Neural Network (INN) that challenges 80 years of standard connectionist dogma.

• The Core Question: Why has artificial intelligence inherited and preserved a crude continuous voltage abstraction ($y=\sigma(Wx+b)$) since 1943, when biological nervous systems compute through finite particle counts, conservation laws, and structural constraints?
• The Mechanism: Instead of real-valued unconstrained activations, computation is carried out through conserved particle counts, ion-specific conductance, receptor stoichiometry, vesicle-mediated output, and local retrograde feedback.
• The Breakthrough: By modeling compartment capacities and mass saturation, a single NIRGEN unit naturally breaks monotonicity and solves the non-linearly separable XOR parity problem without a hidden layer—a physical computation impossible within the voltage abstraction.

---

🏗️ Architecture

A standard artificial neuron computes an unbounded dot-product. NIRGEN models a single ion species ($ion_1$) partitioned across bounded topological sub-domains:

$$\text{Extracellular Cleft }{B}_{ecto}\cup \text{Annular Membrane }B\cup \text{Intracellular Cytosol }{B}_{endo}\backslash text\{Extracellular\; Cleft\; \}\; B\_\{ecto\}\; \backslash ;\backslash cup\backslash ;\; \backslash text\{Annular\; Membrane\; \}\; B\; \backslash ;\backslash cup\backslash ;\; \backslash text\{Intracellular\; Cytosol\; \}\; B\_\{endo\}$$

Problem Input (x) ➔ [Input Patches l_i] ➔ Receptor Gating (r_e) ➔ Ion Translocation
                                                                          │
                                                                   Diffusion Bridge
                                                                          │
Problem Output (y) 📥 [Output Patches l_o] 📥 Vesicle Fusion (V_e) 📥 Intracellular Pool

Technical Specifications

• Geometric Capacities: Intracellular ($C_{B_{endo}}$) and extracellular cleft ($C_{B_{ecto}}$) maximum thresholds determined dynamically by biological cell diameters ($d_{endo}=20\mu\text{m}$, cleft thickness $=20\text{nm}$) and monovalent ion footprints ($\rho \approx 2.92 \text{ ions/nm}^2$).
• Stoichiometric Gating: Receptors require a non-negotiable activation headcount (${r_{l_i}}$) before translocating an integer quantum ($[r_{l_i}]$). Vesicles require an extrusion threshold (${V_{l_o}}$) to release a quantum back into the cleft.
• Trainable Matrix: Optimization updates only the physical receptor inventories ($r_{l_i}$) and readily releasable vesicle pools ($V_{l_o}$).
• Conservation Invariant: $S_{B_{ecto}}^{(t)} + S_{B_{endo}}^{(t)} = D \leq C_{B_{endo}} + C_{B_{ecto}}$ holds exactly across all discrete timestamps $\tau$.

---

🧪 Experimental Matrix

The model was subjected to evaluation across structured parameters to analyze the intersection of physical constraints and computation.

🌐 Group A: Structural Integrity & Geometry

• Experiment 1 — Geometric Scaling Grid: Sweeps homeostatic soma scales ($d_{endo} \in {10, 15, 20, 25},\mu\text{m}$) and perisynaptic cleft profiles ($10\text{nm}$ to $40\text{nm}$) to locate the optimal physical volume for structural stability.
• Experiment 2 — Local Budget Allocation: Evaluates the unit across varying concentration gradients by shifting the ectosphere budget split ($\gamma \in {0.1, 0.25, 0.5, 0.75, 0.9}$) under a frozen total pool.
• Experiment 3 — Free-Floating Control: Removes the localized budget coefficient ($\delta$) to assess computation when cell dynamics are forced to draw unconstrained from the macro ambient pool $P$.

🔍 Group B: Ablations & Relaxations

• Experiment 4 — Gating Hardness Sweep: Tests the convergence performance of backpropagation by annealing the soft-minimum sharpness parameter ($\kappa_{\mathrm{start}} \to \kappa_{\mathrm{end}}$) from smooth approximations down to hard discrete boundaries.
• Experiment 5 — Sub-pool Ratios ($\alpha, \beta$): Ablates the stoichiometric allocation weights that divide ions into independent flux-eligible and ligand-binding pools.
• Experiment 6 — The Continuous Impostor: Replaces the discrete floor boundaries with an unconstrained, continuous real-valued projection of equal parameter width to measure the exact performance decay when physical limits are abandoned.

---

📊 Evaluation & Benchmarks

Calibration and task performance are verified every 50 epochs across three primary paradigms:

1. $n$-Bit XOR Parity: Evaluated via full permutation grids. Tests the network's ability to capitalize on localized compartment saturation to collapse output headroom when co-active inputs collide.
2. Stoichiometric Convergence Rate: Measures parameter trajectory smoothness using Straight-Through Estimators (STE) and Fourier-series smooth floor surrogates against traditional Adam optimization.
3. Mass Invariance Error: A strict conservation audit tracking numerical drift across prolonged forward passes. The acceptable variance bounds are strictly fixed at $0$ due to physical matter laws.

---

💾 Dataset & Training Specs

Data & Resource Grounding

• Biophysical Foundations: Gating rates and ion dimensions are mapped directly from Neher, Sakmann, and MacKinnon's single-channel cryo-EM structural datasets.
• Pharmacological Priors: GPCR desensitization timelines and retrograde attenuation ratios are cross-referenced with empirical Israel/US endocannabinoid clearance kinetics.

Hyperparameters

Parameter	Value
Hardware	Custom Local Environment (Self-Funded Execution)
Optimization Method	Relaxation Backpropagation / STE
Surrogate Operator	$J=10$ Fourier Smooth Floor
Initial Budgets	$D \leq C_{B_{endo}} + C_{B_{ecto}}$ checked dynamically
Mixing Weights	$\sigma = 1.0$ (Membrane Side-Weighting Default)
Learning Rate	0.05 (with strict gradient clipping at norm 5.0)
Sharpness Schedule	Cosine Annealing ($\kappa: 1.0 \to 30.0$)

---

⚠️ Limitations & Disclaimers

• Single Species Approximation: This iteration restricts tracking to a single generic monovalent ion pool ($R={ion_1}$). Real tissue relies on overlapping, highly specific competitive interactions between $\text{Na}^+$, $\le\text{K}^+$, $\text{Cl}^-$, and $\text{Ca}^{2+}$.
• Discrete Step Horizon: The current forward pass resolves across tight, discrete intervals. Fine-grained volume-transmission and asynchronous wave propagation are omitted.
• Instrument Concept Bias Caution: Standard evaluation suites are built to measure vector transformations, not resource allocation. Forcing an INN through standard benchmarks risks falling into the exact Instrument Concept Bias this paradigm fights.

---

📜 Citation

If you utilize this INN paradigm, physical mapping framework, or optimization relaxations, please cite the primary document:

@article{garvey2026shadows,
  title={Shadows of Consciousness: An Investigation into Ion Neural Networks (INN) Using the Neurotransmitter Ion Receptor Glial Endocannabinoid Network (NIRGEN) Paradigm},
  author={Garvey, Myles Douglas},
  journal={Preprint --- Not peer-reviewed. Subject to revision.},
  year={2026}
}

---

Model card written to accompany the formal derivation and calibration of single-unit INN architectures.