Release SONAPSE-ORIGIN v0.2.0: Neural Organisation Discovery, Causal Emergence, PCI_LZ, and CTC
1e6c86d verified | """Experiment R: Precision-Weighted Predictive Coding under Sensory Noise. | |
| Tests whether dynamic precision weighting (attenuating noisy sensory gain and amplifying | |
| top-down prior precision) yields superior sensory reconstruction fidelity under corruption. | |
| """ | |
| from __future__ import annotations | |
| from typing import Any, Dict | |
| import numpy as np | |
| from ..theories.precision_predictive_coding import HierarchicalPredictiveCoder | |
| from .runner import ExperimentRunner | |
| def run_experiment_r(sample_size: int = 15) -> Dict[str, Any]: | |
| runner = ExperimentRunner() | |
| def treatment_dynamic_precision(seed: int) -> float: | |
| coder = HierarchicalPredictiveCoder(layer_sizes=(16, 8, 4), seed=seed) | |
| clean = np.sin(np.linspace(0, 2 * np.pi, 16)) | |
| res = coder.evaluate_noise_robustness(clean, noise_level=0.6, dynamic_precision=True) | |
| fidelity = max(0.0, 1.0 - res["reconstruction_error"]) | |
| return float(fidelity) | |
| def control_static_precision(seed: int) -> float: | |
| coder = HierarchicalPredictiveCoder(layer_sizes=(16, 8, 4), seed=seed) | |
| clean = np.sin(np.linspace(0, 2 * np.pi, 16)) | |
| res = coder.evaluate_noise_robustness(clean, noise_level=0.6, dynamic_precision=False) | |
| fidelity = max(0.0, 1.0 - res["reconstruction_error"]) | |
| return float(fidelity) | |
| return runner.run_controlled_experiment( | |
| exp_id="EXP_R_PRECISION_PREDICTIVE_CODING", | |
| title="Dynamic Precision Weighting Under Sensory Observation Noise", | |
| theory="Predictive-Processing", | |
| statement="Dynamic precision weighting yields significantly higher sensory reconstruction fidelity under severe observation noise compared to static unweighted predictive coding.", | |
| target_metric="reconstruction_fidelity", | |
| treatment_fn=treatment_dynamic_precision, | |
| control_fn=control_static_precision, | |
| sample_size=sample_size, | |
| ) | |