Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
NEURAL NETWORK LEARNING DEVICE, NEURAL NETWORK LEARNING METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2019/208564
Kind Code:
A1
Abstract:
Provided is a neural network learning device capable of adjusting reduction amount of a model size. The neural network learning device comprises a group parameter generating unit that separates model parameters of a neural network model into arbitrarily defined groups and generates group parameters representing features of the respective groups, a regularization term calculating unit that calculates a regularization term under the assumption that distribution of the group parameters conforms to distribution defined by a hyperparameter as a parameter defining the feature of the distribution, and a model updating unit that calculates loss function from a correct-answer label in teacher data, output probability distribution obtained by inputting feature quantity corresponding to the correct-answer label in the teacher data into the neural network model, and the regularization term and updates the neural network model so as to reduce a value of the loss function.

Inventors:
MORIYA TAKAFUMI (JP)
YAMAGUCHI YOSHIKAZU (JP)
Application Number:
PCT/JP2019/017216
Publication Date:
October 31, 2019
Filing Date:
April 23, 2019
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
NIPPON TELEGRAPH & TELEPHONE (JP)
International Classes:
G06N3/08
Other References:
SCARDAPANE, S. ET AL., GROUP SPARSE REGULARIZATION FOR DEEP NEURAL NETWORKS, 2 July 2016 (2016-07-02), pages 1 - 10, XP029946330, Retrieved from the Internet [retrieved on 20190610]
U, X. F. ET AL., BAYESIAN VARIABLE SELECTION AND ESTIMATION FOR GROUP LASSO, vol. 10, no. 4, 3 December 2015 (2015-12-03), pages 909 - 936, XP055648755, Retrieved from the Internet [retrieved on 20190610]
Attorney, Agent or Firm:
NAKAO, Naoki et al. (JP)
Download PDF: