Skip to content

Efficient Unitary Neural Network(EUNN) implementation in Tensorflow

License

Notifications You must be signed in to change notification settings

jingli9111/EUNN-tensorflow

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

87 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EUNN-tensorflow

Unitary neural network is able to solve gradient vanishing and gradient explosion problem and help learning long term dependency. EUNN is an efficient and strictly enforced unitary parametrization based on SU(2) group. This repository contains the implementation of Efficient Unitary Neural Network(EUNN) in tensorflow.

If you find this work useful, please cite arXiv:1612.05231.

I am working on submitting this code to tf.contrib so that in the future you can use it directly from official tensorflow.

Installation

requires TensorFlow > 1.2.0

Demo

./demo.sh

Usage

To use EUNN in your model, simply copy eunn.py.

Then you can use EUNN in the same way you use built-in LSTM:

from eunn import EUNNCell
cell = EUNNCell(hidden_size, capacity, fft, complex)

Args:

  • hidden_size: Integer.
  • capacity: Optional. Integer. Only works for tunable style.
  • fft: Optional. Bool. If True, EUNN is set to FFT style. Default is False.
  • complex: Optional. Bool. If True, EUNN is set to complex domain. Default is True.

Note:

  • For complex domain, the data type should be tf.complex64
  • For real domain, the data type should be tf.float32

Example tasks for EUNN

Copying memory task and pixel-permuted MNIST task for RNN in the paper are shown here. Due to copyright issue, we cannot release TIMIT task.

Copying Memory Task

python copying_task.py --model eunn --T 200 --fft

Pixel-Permuted MNIST Task

python mnist_task.py --model eunn --iter 20000 --hidden 512 --complex False 

License

MIT License

About

Efficient Unitary Neural Network(EUNN) implementation in Tensorflow

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published