Skip to content

Files

Latest commit

cc41e4d · Oct 22, 2015

History

History
9 lines (5 loc) · 423 Bytes

README.md

File metadata and controls

9 lines (5 loc) · 423 Bytes

Deep learning via Hessian-free optimization

(Multi) GPU Implementation of http://machinelearning.wustl.edu/mlpapers/paper_files/icml2010_Martens10.pdf

Trains a neural net using the 2nd order Hessian Free method.

Status of project

The OpenCL kernel for the backward pass is currently an unnecessary bottle-neck. Also, contrastive divergence, dropout and double support are disabled. The code also needs a clean up.