DATA ANALYSIS SYSTEM, LEARNING DEVICE, METHOD, AND PROGRAM
To provide a data analysis system capable of appropriately performing analysis using a CNN model while reducing communication volume.SOLUTION: A data analysis system 90 comprises: a measuring instrument 10 performing conversion processing for outputting compressed data obtained by processing observa...
Saved in:
Main Authors | , , , |
---|---|
Format | Patent |
Language | English Japanese |
Published |
21.05.2020
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | To provide a data analysis system capable of appropriately performing analysis using a CNN model while reducing communication volume.SOLUTION: A data analysis system 90 comprises: a measuring instrument 10 performing conversion processing for outputting compressed data obtained by processing observational data received via an input layer of a learned neural network 18A from the input layer to a predetermined intermediate layer; and an apparatus 20 inputting the compressed data to a next intermediate layer in a learned neural network 18B, and inputting data, which is output from the next intermediate layer, obtained by decrypting the compressed data to an output layer configured using the CNN model, and performing analysis processing in which output of the output layer is result of analyzing the observational data. The learned neural networks 18A and 18B are previously learned so as to minimize an error between the output of the output layer of a true analysis result obtained by analysis when known observational data is input to the input layer and the true analysis result, and an error between the observational data and the decrypted data.SELECTED DRAWING: Figure 4
【課題】通信量を削減しつつ、CNNモデルを用いた適切な分析を行うことができるデータ分析システムを提供する。【解決手段】データ分析システム90は、学習済みニューラルネットワーク18Aの入力層を介して受け付けた観測データを、入力層から所定の中間層まで処理された結果得られる圧縮データを出力する変換処理を行う計器10と、学習済みニューラルネットワーク18Bにおいて圧縮データを次の中間層に入力し、次の中間層の出力である、圧縮データを復号したデータを、CNNモデルを用いて構成された出力層に入力し、出力層の出力を、観測データを分析した結果とする分析処理を行う機器20とを含む。学習済みニューラルネットワーク18A、18Bは、分析して得られる真の分析結果が既知の観測データを入力層に入力したときの出力層の出力と真の分析結果との誤差、及び観測データと復号したデータとの誤差を最小化するように予め学習されている。【選択図】図4 |
---|---|
Bibliography: | Application Number: JP20180209641 |