@techreport{oai:ipsj.ixsq.nii.ac.jp:00210087,
 author = {松本, 麻里 and 古田, 雅則},
 issue = {64},
 month = {Mar},
 note = {秘密データを含むビッグデータの有効活用とプライバシー保護を両立させる方法として，データを秘匿化したまま計算を行う秘密計算の需要が高まっている．近年では機械学習を含むデータ分析に秘密計算を用いる研究が多く行われている．本稿では，Auto-Encoder による異常検知への活用を目的として，平文と同等の検知率を目指した高精度な秘密計算実現に取り組む．秘密分散による秘匿化データを用い，学習および推論では実数を含む秘密計算を実施し，NSL-KDD データセットを用いた評価において，小規模でのノード構成においては平文と同等の約 77% の検知率が得られた．, As a method of achieving both effective utilization of big data including secret data and privacy protection, there is an increasing demand for secret calculation that performs calculations while keeping data confidential. Recently, much research has been done using secret calculations for data analysis, including machine learning. In this paper, we will work on the realization of highly accurate secret computation aiming at the detection rate equivalent to plaintext, with the aim of utilizing it for anomaly detection by the auto-encoder. Using concealed data by secret sharing, secret computation including real numbers is performed in learning and inference, and in evaluation using NSL-KDD data set, about 77% detection equivalent to plain text in small-scale node configuration The rate was obtained.},
 title = {Auto-Encoderによる異常検知手法のための実数を使用した秘密計算方法},
 year = {2021}
}