How star-ga/nn_universe works
This repository is a research-experiment suite for measuring and validating the Fisher Information Matrix (FIM) diagonal tier hierarchy across computational substrates. The main workflow is to run substrate- or architecture-specific measurements, compare tier ratios across systems, and test depth-scaling and statistical separation. Major branches cover neural-network scaling and mechanism studies, non-neural and physics controls, modern pretrained architectures, and statistical analysis. The graph follows the README and file tree; unsampled implementations are represented only at their existing directory paths, with uncertain call wiring omitted.
Main entry points: dichotomy_stats.py Dichotomy statistics · v6_0_depth_mechanism Depth mechanism · v4_0_uniqueness Uniqueness panel · v12_partition_invariant Partition analysis
Components 20
Scaling and mechanism
- Initial scaling
experiments/v1_baseline/scaling_experiment.py - Extended scaling
experiments/v1_2_scaling - Depth mechanism
experiments/v6_0_depth_mechanism - Mechanism controls
experiments/v6_0_mechanism
Substrate experiments
- Uniqueness panel
experiments/v4_0_uniqueness - QEC experiments
experiments/v2_1_qec - Symbolic task
experiments/v3_0_task3_symbolic/symbolic_regression.py - Vision task
experiments/v3_0_task4_vision/vision_classification.py - U(1) lattice
experiments/v5_0_lattice_qcd/lattice_u1.py - SU(2) lattice
experiments/v7_0_lattice_su2/lattice_su2.py - Tensor network
experiments/v8_0_tensor_network
Modern architectures
- Modern architecture panel
experiments/v9_modern_arch - Production baselines
experiments/v10_baselines - MoE measurements
experiments/v11_runpod_pythia_moe
Statistical analysis
- Partition analysis
experiments/v12_partition_invariant - Dichotomy statistics
experiments/v5_0_dichotomy_stats/dichotomy_stats.py - Sensitivity and bootstrap
experiments/v4_3_statistics - Mock α-drift
experiments/v3_1_alpha/mock_pipeline.py - α-drift data stub
experiments/v3_1_alpha/real_data_pipeline.py
Other components
- Researcher
Connections 23
- Researcher → Initial scaling: runs
- Researcher → Extended scaling: runs
- Researcher → Depth mechanism: runs
- Researcher → Uniqueness panel: runs
- Researcher → Modern architecture panel: runs
- Researcher → Production baselines: runs
- Researcher → QEC experiments: runs
- Researcher → U(1) lattice: runs
- Researcher → SU(2) lattice: runs
- Researcher → Tensor network: runs
- Researcher → Symbolic task: runs
- Researcher → Vision task: runs
- Researcher → MoE measurements: runs
- Researcher → Mock α-drift: runs
- Researcher → α-drift data stub: runs
- Researcher → Partition analysis: runs
- Researcher → Sensitivity and bootstrap: runs
- Researcher → Dichotomy statistics: runs
- Depth mechanism → Mechanism controls: tests bounds
- Uniqueness panel → Dichotomy statistics: supplies results
- Dichotomy statistics → Sensitivity and bootstrap: supports inference
- Partition analysis → Dichotomy statistics: tests partitioning
- Mock α-drift → α-drift data stub: informs protocol