Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
INFORMATION PROCESSING DEVICE AND INFORMATION PROCESSING METHOD
Document Type and Number:
WIPO Patent Application WO/2017/073000
Kind Code:
A1
Abstract:
[Problem] To provide an information processing device for efficiently implementing machine learning. [Solution] An information processing device according to one embodiment of the present invention is provided with: an acquisition means for acquiring source code including codes defining a forward process for each layer constituting a neural network; a storage means for storing the correspondence of each forward process with a backward process corresponding to the forward process; and an implementation means for sequentially implementing the codes contained in the source code, the implementation means being configured so that when each code is implemented, the output value of the forward process defined by the code is calculated on the basis of an input value, and a reference structure is generated for the backward process of the layer that corresponds to the code on the basis of the correspondences stored in the storage means.

Inventors:
TOKUI SEIYA (JP)
UNNO YUYA (JP)
OONO KENTA (JP)
OKUTA RYOSUKE (JP)
Application Number:
PCT/JP2016/004027
Publication Date:
May 04, 2017
Filing Date:
September 02, 2016
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
PREFERRED NETWORKS INC (JP)
International Classes:
G06N3/08
Other References:
SEIYA TOKUI, CHAINER NO TSUKAIKATA TO SHIZEN GENGO SHORI ENO OYO, 3 September 2015 (2015-09-03), pages 1 - 40
DOUGAL, MACLAURIN ET AL.: "Grandient-based Hyperparameter Optimization through Reversible Learning", PROCEEDINGS OF THE 32ND INTERNATIONAL CONFERENCE ON MACHINE LEARNING, vol. 37, XP055378774
Attorney, Agent or Firm:
SUDO, Yutaka et al. (JP)
Download PDF: