Title:
LEARNING DEVICE, LEARNING METHOD, AND PROGRAM
Document Type and Number:
WIPO Patent Application WO/2022/176196
Kind Code:
A1
Abstract:
A learning device according to an embodiment of the present invention learns, as latent variables, a probabilistic representation of a target function from input/output data. This learning device is provided with a latent representation estimation unit, a latent representation aggregation unit, a first-level latent variable estimation unit, and a higher-level latent variable estimation unit. The latent representation estimation unit organizes latent variables into a hierarchy and applies input/output data to a neural network to estimate latent representations representing the latent variables. The latent representation aggregation unit aggregates the estimated latent representations for each set of input/output data. The first-level latent variable estimation unit uses the aggregated latent representations to obtain distribution parameters of first-level latent variables among the latent variables in the hierarchy. The higher-level latent variable estimation unit obtains distribution parameters of latent variables at a level higher than the first level.
Inventors:
MIYAHARA MASATO (JP)
SATO DAISUKE (JP)
FUKUDA MASATO (JP)
MATSUMURA NARIMUNE (JP)
NISHIKAWA YOSHIKI (JP)
SATO DAISUKE (JP)
FUKUDA MASATO (JP)
MATSUMURA NARIMUNE (JP)
NISHIKAWA YOSHIKI (JP)
Application Number:
PCT/JP2021/006627
Publication Date:
August 25, 2022
Filing Date:
February 22, 2021
Export Citation:
Assignee:
NIPPON TELEGRAPH & TELEPHONE (JP)
International Classes:
G06N3/08
Domestic Patent References:
| WO2017094267A1 | 2017-06-08 |
Foreign References:
| JP2019075108A | 2019-05-16 |
Other References:
MARTA GARNELO; DAN ROSENBAUM; CHRIS J. MADDISON; TIAGO RAMALHO; DAVID SAXTON; MURRAY SHANAHAN; YEE WHYE TEH; DANILO J. REZENDE; S.: "Conditional Neural Processes", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 4 July 2018 (2018-07-04), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081110939
MAALøE LARS, FRACCARO MARCO, LIéVIN VALENTIN, WINTHER OLE: "BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling", 6 November 2019 (2019-11-06), XP055846375, Retrieved from the Internet [retrieved on 20210930]
MAALøE LARS, FRACCARO MARCO, LIéVIN VALENTIN, WINTHER OLE: "BIVA: A Very Deep Hierarchy of Latent Variables for Generative Modeling", 6 November 2019 (2019-11-06), XP055846375, Retrieved from the Internet
Attorney, Agent or Firm:
KURATA, Masatoshi et al. (JP)
Download PDF: