advanced-weighting-systems

Resonance-based symbolic coupling of heterogeneous neural network models in the AeonLayer

Version 0.1.0 – GenesisAeon Project DOI: 10.5281/zenodo.19110330 Zenodo Record: https://zenodo.org/records/19110330

Resonance-based, symbolically mirrored coupling of heterogeneous neural network models (Transformer, CNN, RNN, GraphNN, Spiking) in the AeonLayer for AeonAI. Built on CREP + Sigillin from unified-mandala.

---

Mathematical Foundation

AeonLayer Aggregation

The central formula of the GenesisAeon stack:

$$ L_{\text{Aeon}} = \sum_i w_i \cdot M_i \cdot \sigma!\left(\beta(R_i - \Theta)\right) $$

Symbol	Meaning
$w_i$	Dynamic resonance weight for model $i$ (from WeightingEngine)
$M_i$	Mirror-Matrix $M_i = P_i \Phi_{\text{sigil}} P_i^\top$ (from SymbolicMirror)
$R_i$	Raw resonance signal for model $i$ (adapter-specific)
$\Theta$	Global resonance threshold
$\beta$	Sharpness / inverse temperature
$\sigma$	Logistic sigmoid

WeightingEngine — Entropy-Governance

Area regime ($S \propto A$):

$$ w_i^{(A)} = \frac{\exp(-\lambda H_i)}{\sum_j \exp(-\lambda H_j)} $$

Volume regime ($S \propto V$, dimension $d$):

$$ w_i^{(V)} = \frac{\exp!\left(-\lambda H_i^{d/2}\right)}{\sum_j \exp!\left(-\lambda H_j^{d/2}\right)} $$

UTAC-Logistic Gate

$$ u_i = \sigma!\left(\kappa(C_i - \tau)\right) $$

CREP — Coherence-Resonance-Entropy Product

$$ \mathrm{CREP}i = \left(1 - \frac{H_i}{H{\max}}\right) \cdot \rho_i \cdot u_i $$

Sigillin Phase Matrix and Mirror-Matrix

$$ M_i = P_i , \Phi_{\text{sigil}} , P_i^\top, \qquad [\Phi_{\text{sigil}}]_{jk} \propto e^{-\frac{1}{2}|j-k|} $$

Resonance Energy

$$ E = | L_{\text{Aeon}} |_F $$

---

Installation

pip install advanced-weighting-systems

Full GenesisAeon Stack

pip install "advanced-weighting-systems[stack]"

Installs: mirror-machine, entropy-governance, sigillin, mandala-visualizer, utac-core, cosmic-web.

Development

git clone https://github.com/GenesisAeon/AdvancedWeightingSystems
cd AdvancedWeightingSystems
pip install -e ".[dev]"

---

Quick Start

import numpy as np
from advanced_weighting_systems.aeon_layer import AeonLayerConfig
from advanced_weighting_systems.symbolic_mirror import (
    TransformerAdapter, CNNAdapter, RNNAdapter, GraphNNAdapter
)
from advanced_weighting_systems.weighting_engine import WeightingConfig
from advanced_weighting_systems.models.coupling import ResonanceCoupling

DIM = 16

coupling = ResonanceCoupling(
    adapters=[
        TransformerAdapter(dim=DIM),
        CNNAdapter(dim=DIM),
        RNNAdapter(dim=DIM),
        GraphNNAdapter(dim=DIM),
    ],
    dim=DIM,
    aeon_config=AeonLayerConfig(beta=1.5, theta=0.0, n_models=4),
    weighting_config=WeightingConfig(lam=1.0),
)

rng = np.random.default_rng(0)
activations = [rng.standard_normal(DIM) for _ in range(4)]
entropies   = rng.uniform(0.1, 0.9, size=4)
coherences  = rng.uniform(0.3, 0.8, size=4)
rho         = rng.uniform(0.4, 1.0, size=4)

output = coupling.step(activations, entropies, coherences, rho)

print(f"L_Aeon shape          : {output.aeon_state.layer_output.shape}")
print(f"Resonance energy      : {output.energy:.6f}")
print(f"CREP scores           : {output.weighting_result.crep_scores}")

---

CLI

# Couple four adapter types with AeonLayer + mandala ASCII visualisation
aws couple --models trans,cnn,rnn,graph --aeon-layer --entropy 0.4 --visualize

# Compute resonance weights for named models with CREP bar chart
aws weight --model-ids model_a,model_b,model_c --regime volume --lam 2.0 --visualize

---

Architecture

advanced_weighting_systems/
├── aeon_layer.py        # L_Aeon aggregation (AeonLayer, AeonLayerConfig)
├── weighting_engine.py  # Entropy-governance, UTAC, CREP (WeightingEngine)
├── symbolic_mirror.py   # Sigillin + MandalaMap -> M_i (SymbolicMirror, adapters)
├── cli.py               # Typer CLI (aws couple / aws weight)
├── models/
│   ├── __init__.py      # Re-exports all adapters
│   └── coupling.py      # ResonanceCoupling end-to-end pipeline
└── utils/
    ├── __init__.py
    └── entropy.py       # Shannon entropy, EntropyTable, governance weights

Supported NN Adapters

Adapter	--models key	Resonance Signal
TransformerAdapter	trans	mean(tanh(activations))
CNNAdapter	cnn	max(abs(activations))
RNNAdapter	rnn	std(activations)
GraphNNAdapter	graph	mean(activations^2)
SpikeAdapter	spike	spike rate above threshold

---

Testing

pytest                          # run all tests with coverage
pytest --cov-report=html        # HTML coverage report

Coverage target: 99 %+ (enforced via --cov-fail-under=99).

Contract tests verify interface compatibility with:

• mirror-machine — Mirror-Matrix shape and finite-value contracts
• entropy-governance — monotonicity of weights w.r.t. entropy
• utac-core — UTAC gate range and monotonicity w.r.t. coherence

---

Documentation

mkdocs serve          # local preview
mkdocs build --strict # production build (zero warnings)

Published at: https://genesisaeon.github.io/AdvancedWeightingSystems

---

Citation

@software{genesisaeon_aws_2026,
  author    = {GenesisAeon},
  title     = {Advanced Weighting Systems: Resonance-based Coupling of Heterogeneous NN Models},
  year      = {2026},
  version   = {0.1.0},
  doi       = {10.5281/zenodo.19110330},
  url       = {https://github.com/GenesisAeon/AdvancedWeightingSystems}
}

---

License

Dual-licensed:

• Code (src/, tests/, scripts, configuration) — GPL-3.0-or-later
• Documentation (README, docs/, RELEASE_GUIDE.md, CONTRIBUTING.md, CHANGELOG.md, templates) — CC BY 4.0

See LICENSE for the full split. — GenesisAeon Project, 2026.