DATA META-SCALING APPARATUS AND METHOD FOR CONTINUOUS LEARNING

PROBLEM TO BE SOLVED: To optimize the contraction reference for data contraction by sustained knowledge enhancement in various dimensions that can represent data in the course of machine learning execution.SOLUTION: The apparatus includes a meta optimizer 10 for setting contraction reference information defining a rule to contract input data so as to be expressed by other attributes, learning reference information defining a rule to limit learning to the contracted data and an evaluation rule of learning performance, and knowledge enhancement reference information defining a rule for optimizing the contraction reference information, a contractor 20 for contracting the input data into contracted data based on the contraction reference information, a learning machine 30 for generating a learning model by performing learning on the contracted data based on the learning reference information, and an evaluator 40 for evaluating the performance of the learning model based on the learning reference information and determining the suitability of the contracted data. The meta optimizer performs knowledge enhancement for updating the contraction reference information based on the knowledge enhancement reference information according to a determination result of the suitability.SELECTED DRAWING: Figure 1 【課題】機械学習の実行過程においてデータを表現できる様々な次元で持続的な知識増強によってデータ縮約のための縮約基準を最適化する。【解決手段】入力データを他の属性で表現されるように縮約する規則を定義した縮約基準情報、前記縮約データへの学習を制限する規則と学習性能の評価規則を定義した学習基準情報及び縮約基準情報を最適化するための規則を定義した知識増強基準情報を設定するメタ最適化機１０と、縮約基準情報に基づいて入力データを縮約データに縮約する縮約機２０と、学習基準情報に基づいて縮約データへの学習を行なって学習モデルを生成する学習機３０と、学習基準情報に基づいて前記学習モデルの性能を評価して、縮約データの適切性を判断する評価機４０とを含む。メタ最適化機は前記適切性を判断した結果に応じて前記知識増強基準情報に基づいて前記縮約基準情報を更新する知識増強を行う。【選択図】図１