-
-
Notifications
You must be signed in to change notification settings - Fork 0
Home
MASELKO-95 edited this page Nov 28, 2025
·
1 revision
Welcome to the Collatz AI Research System documentation!
- Getting Started - Installation and first run
- Model Architecture - Deep dive into the neural network
- Training Guide - How to train and optimize
- Loop Searcher - Brute-force cycle detection
- Performance Analysis - Metrics and results
- API Reference - Code documentation
- FAQ - Common questions
This project combines AI-guided pattern recognition with parallel brute-force search to investigate the Collatz Conjecture.
| Metric | Value |
|---|---|
| Training Steps | 100,000+ |
| Numbers Analyzed | 54,000,000+ |
| Model Accuracy | 99.97% (stopping time) |
| GPU Utilization | 90% |
| CPU Utilization | 85% |
- ✅ No non-trivial cycles found in range [2^68, 2^68 + 54M]
- 📈 Stopping time prediction: 0.0003 log-space error
- 🎯 Sequence accuracy: ~70% per-step prediction
git clone https://github.com/MASELKO-95/Collatz-AI-Research-System.git
cd Collatz-AI-Research-System
chmod +x run.sh
./run.shThis wiki is organized into the following sections:
- Getting Started - Setup, installation, and first run
- Architecture - Technical details of the AI model
- Training - How to train and customize
- Analysis - Understanding results and metrics
- Advanced - Distributed training, optimization tips
- Contributing - How to help improve the project
We welcome contributions! See the Contributing Guide for details.
- Issues: GitHub Issues
- Discussions: GitHub Discussions
Last updated: 2025-11-28
Welcome to the Collatz AI Research System documentation!
- Getting Started - Installation and first run
- Model Architecture - Deep dive into the neural network
- Training Guide - How to train and optimize
- Loop Searcher - Brute-force cycle detection
- Performance Analysis - Metrics and results
- API Reference - Code documentation
- FAQ - Common questions
This project combines AI-guided pattern recognition with parallel brute-force search to investigate the Collatz Conjecture.
| Metric | Value |
|---|---|
| Training Steps | 100,000+ |
| Numbers Analyzed | 54,000,000+ |
| Model Accuracy | 99.97% (stopping time) |
| GPU Utilization | 90% |
| CPU Utilization | 85% |
- ✅ No non-trivial cycles found in range [2^68, 2^68 + 54M]
- 📈 Stopping time prediction: 0.0003 log-space error
- 🎯 Sequence accuracy: ~70% per-step prediction
git clone https://github.com/MASELKO-95/Collatz-AI-Research-System.git
cd Collatz-AI-Research-System
chmod +x run.sh
./run.shThis wiki is organized into the following sections:
- Getting Started - Setup, installation, and first run
- Architecture - Technical details of the AI model
- Training - How to train and customize
- Analysis - Understanding results and metrics
- Advanced - Distributed training, optimization tips
- Contributing - How to help improve the project
We welcome contributions! See the Contributing Guide for details.
- Issues: GitHub Issues
- Discussions: GitHub Discussions
Last updated: 2025-11-28