MULTIPLEX LEARNING SYSTEM AND MULTIPLEX LEARNING PROGRAM

PROBLEM TO BE SOLVED: To ensure that output is the same as before regardless of version upgrade of a leaning device.SOLUTION: A plurality of learning devices 2a-2c accept input data made up of m-dimensional elements as input and provide output data made up of n-dimensional elements on the basis of a...

Full description

Saved in:
Bibliographic Details
Main Author ADACHI MASAHIKO
Format Patent
LanguageEnglish
Japanese
Published 18.10.2018
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:PROBLEM TO BE SOLVED: To ensure that output is the same as before regardless of version upgrade of a leaning device.SOLUTION: A plurality of learning devices 2a-2c accept input data made up of m-dimensional elements as input and provide output data made up of n-dimensional elements on the basis of a prescribed function including pre-adjusted internal parameters, where m is a natural number of m≥2 and n<m. The learning devices 2a-2c are added in time-sequence order, and their functions or values of internal parameters are different from each other. A past record database 6 stores each piece of actual record data including input data adopted as past actual records in association with one learning device (any of the learning devices 2a-2c) which processed the actual record data. A selection unit 3 refers to storage content of the actual record database 6, and selectively outputs any of output data from the learning devices 2a-2c as output of input data which is a processing object this time.SELECTED DRAWING: Figure 1 【課題】学習器のバージョンアップに関わりなく、従前の出力との同一性を確保する。【解決手段】複数の学習器2a〜2cは、予め調整された内部パラメータを含む所定の関数に基づいて、m(m≧2の自然数)次元の要素よりなる入力データを入力として、n(n<m)次元の要素よりなる出力データを出力する。これらの学習器2a〜2cは、時系列的な順序で追加され、それぞれの関数または内部パラメータの値が互いに異なる。実績データベース6は、過去の実績として採用された入力データを含む実績データのそれぞれを、この実績データの処理を行った一の学習器(学習器2a〜2cのいずれか)と対応付けて記憶する。選択部3は、実績データベース6の記憶内容を参照して、今回の処理対象となる入力データの出力として、複数の学習器2a〜2cの出力データのいずれかを選択的に出力する。【選択図】図1
Bibliography:Application Number: JP20170062012