Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,47 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: other
|
| 3 |
-
license_name: other
|
| 4 |
-
license_link: LICENSE
|
| 5 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: other
|
| 4 |
+
license_link: LICENSE
|
| 5 |
+
---
|
| 6 |
+
---
|
| 7 |
+
tags:
|
| 8 |
+
- green-ai
|
| 9 |
+
- edge-computing
|
| 10 |
+
- c++
|
| 11 |
+
- spectral-graph-theory
|
| 12 |
+
- ramanujan-graphs
|
| 13 |
+
- topological-deep-learning
|
| 14 |
+
license: other
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# ⚡ Ramanujan Spectral Reservoir (RSR)
|
| 18 |
+
|
| 19 |
+
> **"Intelligence is not about weight adjustment, but optimal topology."**
|
| 20 |
+
|
| 21 |
+
This repository hosts the reference implementation and benchmarks for the **Ramanujan Spectral Reservoir**, a topological AI architecture that replaces Backpropagation with closed-form solutions on spectral expander graphs.
|
| 22 |
+
|
| 23 |
+
## 🚀 Key Benchmarks
|
| 24 |
+
We achieved **hard real-time** performance on commodity hardware by eliminating iterative training in hidden layers:
|
| 25 |
+
|
| 26 |
+
| Device | Metric | Result | Speedup vs MLP |
|
| 27 |
+
| :--- | :--- | :--- | :--- |
|
| 28 |
+
| **Legacy CPU (i5-4570, 4th Gen)** | Inference Time | **~287x Faster** | 287x |
|
| 29 |
+
| **Mobile (Android ARM64)** | Latency | **< 0.6 ms** | >100x |
|
| 30 |
+
| **Throughput** | FPS | **1600+ inf/sec** | N/A |
|
| 31 |
+
|
| 32 |
+
## 📄 The Paper
|
| 33 |
+
Full mathematical derivation, proofs, and the "Poliform Industrial Secret Protocol" details are available on Zenodo:
|
| 34 |
+
**[LINK A TU ZENODO AQUÍ]**
|
| 35 |
+
|
| 36 |
+
## 🔧 How it Works
|
| 37 |
+
1. **Project:** Input data is projected onto a fixed **Ramanujan Graph** ($d$-regular spectral expander).
|
| 38 |
+
2. **Diffuse:** Information propagates via spectral diffusion (mixing time is optimal).
|
| 39 |
+
3. **Solve:** The readout layer is computed analytically (Closed-Form) using Ridge Regression.
|
| 40 |
+
4. **Result:** Deterministic, Green AI that runs on bare metal C++.
|
| 41 |
+
|
| 42 |
+
## 💻 Code Availability
|
| 43 |
+
The C++ kernel for Android and the Python reference implementation are available under the **Polyform Strict License 1.0.0** (Non-commercial research only).
|
| 44 |
+
|
| 45 |
+
---
|
| 46 |
+
*Developed by Andrés Sebastián Pirolo (Independent Researcher).*
|
| 47 |
+
*Contact: apirolo@abc.gob.ar*
|