A Novel Machine Learning Framework Combining Meta-Learning and Q-Learning with Pathfinding Optimization
MetaQ-Star is a pioneering machine learning framework that combines the power of meta-learning with reinforcement learning, specifically Q-learning, enhanced by optimized pathfinding algorithms. This unique integration enables models to adapt quickly to new tasks with minimal data while efficiently exploring solution spaces.
MetaQ-Star introduces a novel approach where meta-learning principles enable rapid adaptation across tasks, while Q-learning provides robust reinforcement learning capabilities. The framework's pathfinding component optimizes the exploration of complex solution spaces, leading to faster convergence and better generalization.
- Meta-Learning Engine: Implements newly developed Mode-Conditional Bayesian Model-Agnostic Meta-Learning framework for fast adaptation to new tasks
- Q-Learning System: Advanced reinforcement learning implementation using a factorized, hierarchical double-agent Q-Learning algorithm
- Pathfinder: A* inspired optimization using a bi-directional and diagonal search algorithm for efficient solution space exploration in Q-tables
- Hyperparameter Optimization: Hyperparameter tuning and optimization using Optuna
- Distributed Computation: Ray integration for scaling across computational resources
- Cache Management: Intelligent caching system for optimized memory usage
- Comprehensive Logging: Detailed tracking of experiments and model performance
- Database Integration: Configurable database solutions for different environments
MetaQ-Star is built on a modern Python stack:
- Python 3.12
- PyTorch 2.6
- CUDA 12.6
- Pydantic V2
- Optuna v4.2.1
- Ray 2.43.0
- Multi-device support: CUDA (preferred), MPS (secondary), CPU (backup)
- Cloud support for using MongoDB and Redis
- Local support for using SQLite3 and system memory
For detailed documentation, please refer to the research_docs directory which includes implementation details and theoretical background.
If you use MetaQ-Star in your research, please cite:
@software{metaq_star2023,
author = saviornt,
title = MetaQ-Star: A Novel Machine Learning Framework,
year = 2025,
url = {https://github.com/saviornt/MetaQ-Star}
}
We welcome contributions! Please see CONTRIBUTING.md for details on how to submit pull requests, report issues, and join the development effort.