You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Brain-5D

Sparse 5D spiking-neural research framework with controlled plasticity, embodiment, peripheral multi-network integration and scientific provenance.

Brain-5D is an experimental framework for studying learning, self-organization and embodied closed-loop behavior in a sparse five-dimensional spiking neural network. The SNN remains the primary adaptive system; language models, research assistants and peripheral neural networks are bounded interpretation/proposal/adapter components unless explicitly registered as experimental treatments.

Brain-5D does not claim AGI, consciousness, sentience or biological equivalence. Passing implementation tests, reachable devices or available neural pipelines are not the same as scientific evidence.

Current baseline — 2026-09-07

version: 0.5.0a7
Python: 3.11–3.13
latest verified Python suite: 825 passed, 5 skipped
latest verified browser suite: 5 passed with Chromium
Research Catalog / variable-projection-dimension integration is merged to main
merge commit: 85e7209509b348bf7912dde01d3d9ebb078a2e61
latest fully completed pre-merge main CI baseline: run #598, success
the post-merge CI run is authoritative for the merged baseline and must complete before that baseline is described as fully green
GitHub main is canonical; this file is published as the Hugging Face repository README during mirror synchronization

Live dashboard Space

The interactive operator and research dashboard is published separately as a Docker Space:

https://huggingface.co/spaces/superdigger/Brain-5D-Space

Capabilities

sparse 5D Izhikevich SNN with delayed events and deterministic RNG state;
STDP, signed eligibility and delayed reward / three-factor learning;
homeostasis and bounded interoception;
structural proposals, explicit approval, mutation, journal, undo and recovery;
.b5d snapshots, delta journals, checkpoints and deterministic restore/continue;
typed embodiment, actuator authorization, audit trails and deterministic environments;
real host telemetry/device discovery without fabricated fallback values;
Neural Symbiosis: open-set peripheral neural/virtual area contracts and read-only pipeline reachability;
MSBA: modality-specific pathways, energy/resource accounting and external projection dimensionality from 1–32 dimensions;
fragmentable canonical research-question/hypothesis registries with duplicate-ID rejection;
searchable Research Catalog with operational/exploratory distinction;
repository-wide read-only RQ/H reference audit;
research registries, manifests, DATA/EVID separation and AI provenance;
responsive dashboard centered on Overview, Control, Research, Settings, Wesen and Embodiment.

Research Catalog and dimensions

Canonical questions.yaml / hypotheses.yaml can be extended through deterministic questions.*.yaml / hypotheses.*.yaml fragments. MSBA research questions and hypotheses are normal experiment-workflow entries. A question is marked operational only when an appropriate frozen/preregistered protocol exists; otherwise it remains exploratory and cannot be silently promoted to evidence.

MSBA/external projection spaces may use 1–32 dimensions. The persisted productive SNN core remains 5D for backward compatibility until a separately versioned N-D neuron-ID, spatial-index and .b5d migration has been implemented and verified.

Neural Symbiosis

The embodiment layer can represent dedicated processing stages between endpoints and the 5D-SNN:

Camera → CNN / Vision Transformer → gateway → 5D-SNN
Microphone → Audio/Speech Transformer → gateway → 5D-SNN
Database / Knowledge Graph → GNN/projector → gateway → 5D-SNN
Logic engine → neuro-symbolic projector → gateway → 5D-SNN
5D-SNN → gateway → speech/control network → audio/robotics output

The adapter model is open-set and framework-neutral. CNN, Transformer, LSTM/GRU/RNN, GNN, Modern Hopfield, reservoir/ESN, MLP, VAE/GAN/diffusion, autoencoder, peripheral SNN, multimodal, neuro-symbolic and custom architectures can be represented without importing their runtime frameworks into the Brain-5D core.

Gateway plasticity is disabled by default. Pipeline reachability or area registration is not evidence that the SNN learned to use an external area. Plastic gateway experiments require explicit preregistration, persisted RNG/model/version provenance, matched controls and the normal DATA/EVID review path.

Quick start

git clone https://github.com/Thomas-Heisig/Brain-5D.git
cd Brain-5D
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
python -m src.main --config configs/poc_config.yaml

Windows PowerShell activation:

.\.venv\Scripts\Activate.ps1
pip install -e ".[dev]"
.\start.ps1

The dashboard defaults to http://127.0.0.1:8765.

Current research focus

The engineering foundation is broad enough that the next priority is evidence closure rather than feature accumulation:

operationalize every still-unmapped canonical RQ/H with dedicated protocols and preregistrations;
post-repair propagation/recurrence validation across independent seeds;
productive learning with holdout and matched controls;
closed-loop embodiment versus replay/open-loop controls;
experiment-only Neural Symbiosis/MSBA gateway studies with frozen/random/shuffled controls;
preregistered N-D projection sweeps before any productive-core N-D migration;
simulation-time versus wall-clock pacing calibration;
causal ablations of the 5D organization;
self-regulation/sensor-loss studies;
later memory/world-model, multimodal grounding and AI-as-treatment studies.

Documentation

Versioned Alpha/Sprint/Release documents are historical traceability records and should not be used as the current project status unless linked by a canonical document.

Scientific boundary

implementation test != experiment data != accepted evidence != interpretation

Observed values remain distinct from inferred values. Missing telemetry remains unknown. Device availability is not authorization. AI output is not empirical measurement. Pipeline reachability is not learned tool use.

License and citation

MIT License — see LICENSE.

@software{heisig2026brain5d,
  author  = {Thomas Heisig},
  title   = {Brain-5D: Sparse 5D Spiking-Neural Research Framework},
  year    = {2026},
  version = {0.5.0a7},
  url     = {https://github.com/Thomas-Heisig/Brain-5D},
  license = {MIT}
}

GitHub: https://github.com/Thomas-Heisig/Brain-5D

Downloads last month: -; Downloads are not tracked for this model. How to track

Video Preview

Reinforcement Learning