Keras implementation of Efficient Neural Architecture Search

• ALMOST DONE: CNN micro search implementation (now testing)
• TODO: RNN cell search
• TODO: CNN macro search

• Python 3.6+
• Keras 2.1.5+
• Tensorflow 1.0.1+

• GPU

Run either ENAS_Keras_CIFAR10.py or ENAS_Keras_CIFAR10.ipynb on Jupyter Notebook to micro search CNN cells using Cifar10 dataset. Run either ENAS_Keras_MNIST.py or ENAS_Keras_MNIST.ipynb on Jupyter Notebook to micro search CNN cells using MNIST dataset.

.
├── ENAS_Keras_CIFAR10.ipynb
├── ENAS_Keras_CIFAR10.py
├── ENAS_Keras_MNIST.ipynb
├── ENAS_Keras_MNIST.py
├── ENAS.py
├── __init__.py
└── src
    ├── child_network_micro_search.py
    ├── controller_network.py
    ├── __init__.py
    ├── keras_utils.py
    └── utils.py

• Tensorflow
• Pytorch

• Neural Architecture Search with Reinforcement Learning
• Learning Transferable Architectures for Scalable Image Recognition
• Efficient Neural Architecture Search via Parameter Sharing