@techreport{oai:ipsj.ixsq.nii.ac.jp:00241192,
 author = {田中, 哲士 and 須藤, 弘貴 and 岸, 寿春 and 森田, 哲之 and Satoshi, Tanaka and Hiroki, Sudo and Toshiharu, Kishi and Tetsushi, Morita},
 issue = {4},
 month = {Nov},
 note = {データを暗号化したまま分析を行う秘密計算技術は機微なデータの安全な分析や企業間の横断分析への利用が期待でき，近年では AI モデルの学習・予測を秘密計算上で実現する秘密計算 AI の研究が進められている．将来的には秘密計算による大規模なデータ学習が期待されるが，秘密計算は暗号化する性質から処理時間，メモリ消費量が増大し現状は困難である．本研究では，秘密計算 AI を高速化するためにデータ削減方式を検討し，削減方式における処理時間，メモリ消費量の削減効果，精度への影響について述べる．, The secure computation technology that performs analysis while keeping the data encrypted is expected to be used for secure analysis of sensitive data and cross-company analysis, and in recent years, research on secure computation AI that realizes AI model learning and prediction on secure computation has been progressing. In the future, it is expected that large-scale data learning will be possible using secret computation, but at present it is difficult due to the nature of secret computation, which involves encryption, and increases processing time and memory consumption. In this study, we examine data reduction methods to accelerate secret computation AI, and discuss the reduction effects on processing time and memory consumption in the reduction method, as well as the impact on accuracy.},
 title = {秘密計算AIの実用化に向けたデータ削減手法の検討},
 year = {2024}
}