You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
||4 years ago|
|examples||4 years ago|
|images||4 years ago|
|.gitignore||4 years ago|
|LICENSE||4 years ago|
|README.rst||4 years ago|
|dni.py||4 years ago|
|setup.py||4 years ago|
Decoupled Neural Interfaces for PyTorch
This tiny library is an implementation of
`Decoupled Neural Interfaces using Synthetic Gradients <https://arxiv.org/abs/1608.05343>`_
for `PyTorch <http://pytorch.org/>`_.
It's very simple to use as it was designed to enable researchers to integrate
DNI into existing models with minimal amounts of code.
To install, run::
$ python setup.py install
Description of the library and how to use it in some typical cases is provided
below. For more information, please read the code.
This library uses a message passing abstraction introduced in the paper. Some
terms used in the API (matching those used in the paper wherever possible):
- ``Interface`` - A Decoupled Neural Interface that decouples two parts (let's
call them part A and part B) of the network and lets them communicate via
``message`` passing. It may be ``Forward``, ``Backward`` or
- ``BackwardInterface`` - A type of ``