mehdiabdu391-gif/Synapse

Lightweight offline adaptive pattern recognition engine. Runs on any Android phone. No internet, no cloud, no GPU required. Built solo in Ethiopia. # SYNAPSE — Self-Organizing Hierarchical Pattern Recognition Engine **Built by Mehdi Abdu Mohammed — Solo, Ethiopia, on a $80 Android phone.** --- ## What is SYNAPSE? SYNAPSE is a lightweight, offline, adaptive pattern recognition engine that runs on the cheapest Android hardware available. It learns continuously from streaming data, organizes itself, prunes what it doesn't need, and works without any internet connection, cloud service, or GPU. It was built alone, tested on a Samsung A03 Core and A10s using Pydroid 3, and validated under extreme stress conditions. If it runs there, it runs anywhere. --- ## Why does this exist? Most machine learning systems require: - Cloud connectivity - High-end hardware - Pre-labelled datasets - Retraining when data changes SYNAPSE requires none of these. It was built with one constraint in mind: **work on the cheapest phone possible, offline, in real time.** That constraint makes it useful everywhere else too. --- ## What makes it different? SYNAPSE unifies six established mathematical frameworks into a single deployable engine — something that hasn't been packaged this way before: - **Robbins-Monro Stochastic Approximation** — provably convergent online learning - **Manifold Regularization** — geometric alignment via graph Laplacian - **Adaptive Resonance Theory** — novelty detection and stability - **Dirichlet Process Mixture Models** — Bayesian nonparametric node creation - **Minimum Description Length (MDL)** — optimal pruning of redundant structure - **Birth-Death Edge Dynamics** — sparse, self-organizing connectivity No single existing system (SOM, GNG, ART, online DPMM) combines all of these into one lightweight, offline-deployable engine. --- ## Hardware it runs on | Device | RAM | Processor | Performance | |--------|-----|-----------|-------------| | Samsung A03 Core | 2GB | MediaTek | ~360 steps/sec | | Samsung A10s | 2GB | Exynos 7884 | ~850 steps/sec | Memory usage: ~45MB for 1000 data points CPU pe …