LEARNING QUALITY ESTIMATION DEVICE, METHOD, AND PROGRAM

• Main Authors: NAGATA MASAAKI, MORISHITA MUTSUMI, SUZUKI JUN
• Format: Patent
• Language: English; Japanese
• Published: 05.09.2019
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

To provide a learning quality estimation device, a method, and a program capable of removing erroneous data from learning data used for machine learning such as natural language processing.SOLUTION: A learning quality estimation device 90 includes a storage unit 16 for storing, based on a plurality of first learning pairs in which a first discrete sequence indicating an input of a discrete sequence and a second discrete sequence indicating an output have a correct correspondence relationship, a forward-trained model of a discrete sequence converter that converts the first discrete sequence of a discrete sequence into the second discrete sequence that has been learned in advance; and a quality score calculation unit 14 that calculates a quality score using a forward-learned model for a second learning pair that includes a discrete sequence input and an output that may include an error in the correspondence.SELECTED DRAWING: Figure 1 【課題】自然言語処理等の機械学習に用いる学習用データ中の誤ったデータを取り除くことができる学習品質推定装置、方法、及びプログラムを提供する。【解決手段】学習品質推定装置９０は、離散系列の入力を示す第１離散系列と出力を示す第２離散系列とが正しい対応関係である複数の第１学習用ペアに基づいて予め学習された、離散系列の第１離散系列を第２離散系列に変換する離散系列変換器の順方向学習済みモデルを記憶した記憶部１６と、対応関係に誤りを含んでいる可能性のある離散系列の入力及び出力からなる第２学習用ペアについて、順方向学習済みモデルを用いて品質スコアを計算する品質スコア計算部１４と、を備える。【選択図】図１