rmems/Spikenaut-SNN

Gemini said Spikenaut-SNN-v2: Neuromorphic Reservoir & Hybrid Architectures ⚠️ Research Status: Ground Zero Rebuild (April 2026) I am currently taking this repository through a complete architectural reset. Following my initial prototype phase, I have moved back to a "Ground Zero" state to ensure absolute verification of every spiking neuron, synapse, and temporal parameter.

I realized I was moving too fast after my first successful prototype. To maintain the integrity of my research, I am now proceeding at a steadier, more deliberate pace. My current focus is centered on verifying telemetry integrity and ensuring my hardware-software baseline is bulletproof before scaling back into complex hybrid systems.

🧠 Model & Research Description Spikenaut-SNN-v2 is my research-grade Spiking Neural Network (SNN) model designed for high-frequency data processing and autonomous decision-making in noisy environments.

Core Research Objectives: Hardware-Driven Learning: Utilizing real-world telemetry from hardware—specifically my Dynex mining baseline—to drive STDP (Spike-Timing-Dependent Plasticity).

Hybridization (SpikeLMo): Investigating the infusion of SNN logic into the OLMoE-7B (Mixture of Experts) architecture. I am exploring the use of SNNs as low-power, temporal "Neuromorphic Routers" to gate high-level LLM experts.

HFT Logic: Developing specialized 16-neuron Liquid State Machine (LSM) reservoirs tuned for high-frequency trading market micro-structures.

🏗 System & Infrastructure This research is developed and trained locally on my Ship of Theseus workstation running Fedora 43. My codebase is organized under the Eagle-Lander framework.

Primary Logic: Powered by my neuromod Rust crate.

Methodology: "Measure twice, spike once." I have moved away from bulk AI-assisted data uploads to a manual, deterministic verification process to eliminate noisy or "bad" data.

Hardware Integration: The Digilent Basys 3 FPGA.

🛠 Project Roadmap [x] Initial Prototype (Dynex SNN)

[ ] CURRENT: Telemetry verification and baseline stabilization.

[ ] 16-neuron LSM reservoir tuning for HFT data.

[ ] Prototype integration with OLMoE-7B MoE layers.

📜 License This model and its associated logic are released under the GNU General Public License v3.0 (GPL-3.0).

This documentation and research summary were drafted by Gemini, a large language model built by Google, based on the specific research parameters and project history provided by Raul Montoya Cardenas