Study on fast and accurate classifier learning method from unlabeled big data

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K21815
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Review Section: Medium-sized Section 61:Human informatics and related fields
• Research Institution: Osaka University
• Principal Investigator: Washio Takashi 大阪大学, 産業科学研究所, 教授 (00192815)
• Project Period (FY): 2020-07-30 – 2024-03-31
• Keywords: 弱教師有り学習 / 分類器学習 / 機械学習 / UUC / 教師ラベル無しデータ

Outline of Final Research Achievements

With the widespread adoption of AI technology, there is an increasing demand for classifier learning from unlabeled big data due to constraints and costs associated with data collection. In response to this issue, the UUC method, which learns classifiers from two unlabeled datasets with different proportions of positive and negative examples, has been proposed. However, existing methods require vast computational resources for large-scale data and suffer from bias error in classification.
In this study, we propose a versatile UUC method which requires low computational cost only, and is free from bias error. We applied this method to the classification of various datasets, including real data, and verified that unsupervised learning without teaching labels is possible with almost the same accuracy as supervised learning. This establishes a UUC method that far exceeds the application range limitations of the existing UUC methods.

Free Research Field

人工知能

Academic Significance and Societal Importance of the Research Achievements

IoT社会の深化とAI技術の普及に伴い、ビッグデータからの分類器学習ニーズが増しているが、多くの場合にデータ収集の制約やコストから教師ラベルが得られないことが問題となっている。これに対し近年、正負例割合の異なる２つのラベル無し事例集合から分類器を学習するUUC手法が提案されている。しかし、これらはカーネル法を用いており、訓練データ数NについてO(N3)の学習計算量を要し、またN→∞でも分類に偏り誤差を生じる場合がある。従って、複雑な事例分布を持つビッグデータに適用可能な高速高精度なUUC手法の開発が強く待たれていた。本研究成果は、この社会的要請に応えるものである。