Neural Architecture Search
PyTorch
Ray
CUDA
An automated machine learning framework for discovering optimal neural network architectures using reinforcement learning algorithms. This project implements a scalable approach to architecture search with distributed training.
Features
- Policy-based reinforcement learning for architecture design
- Distributed search space exploration with Ray
- GPU acceleration with CUDA optimizations
- Efficient candidate evaluation through shared weights
- Progressive architecture growth to manage complexity
- Automated hyperparameter tuning