Skip to main content

Hitachi

Corporate InformationResearch & Development

June 13, 2017

Report from Presenter

International conference GISTAM2017 was held at Porto, Portugal from Apr. 26th 2017 to 29th in 3days. The author presented our technology to maintain spatial distribution data there. The emerging conference, which was the third time in 2017, is relevant to geospatial data processing technologies. Many researchers mainly coming from Europe presented their research results.


Fig. 1 Usecase Illustration
Enlarge

The author's presentation was focused on our proposed techniques to store various distribution data (such as rainfall etc.) into a relational database. A general database management system has difficulties to maintain the data due to numerous records for the data. The proposed techniques contribute to reduce the number of records without loss of convenience to use, encouraging use of the distribution data as a "Bigdata". The proposed techniques, additional tricks for accelerating the reduction processes, and the added experimental results were summarized in the presentation at the conference.

A feature of the proposed techniques is reduction with kernel regression, which is one of the machine learning methods used for prediction. Even if stored distribution data are partial, the original data can be restored in small errors with prediction based on kernel regression. Attending the kernel regression method, we successfully developed a method that the restoration errors can be constrained to keep less than a given threshold. Though one of the behinds of the proposed techniques is long processing time, details of acceleration with a heap tree and performance varied with parameter settings were shown in the presentation.


Fig. 2 Precision comparison among parameters
Enlarge


Fig. 3 Sparse data (only red triangles are stored for reconstruction)
Enlarge

The proposed techniques support analysis of quantity distribution data that has difficulties about analytics, being expected to lead new values. We will provide more techniques to produce more values, for the future society using much values produced with various data.

  • Page top