Skip to content

Latest commit

 

History

History
19 lines (17 loc) · 741 Bytes

README.md

File metadata and controls

19 lines (17 loc) · 741 Bytes

DCDFA

BriCA2 code tweaked for Decoupled Direct Feedback Alignment (DCDFA) training.

Usage

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -Dtest=true ..
make
./src/examples/dcdfa/dcdfa

About DCDFA

Decoupled direct feedback alignment is an algorithm for training deep neural networks. Each layer is trained in a semi-supervised fashion by autoencoding each input fed and performing direct feedback alignment using errors fed back from the output layer. Individual layers can be trained in a concurrent manner thus 'decoupled' from the rest of the network. An example included in this repository demonstrates the parallel training of a 100 layer fully-connected neural network implemented atop the BriCA2 framework.