Development of Discriminative Pattern Mining Techniques as a Foundation of Human-Centric Machine Learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K11941
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Meijo University
• Principal Investigator: Kameya Yoshitaka 名城大学, 情報工学部, 准教授 (60361789)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 識別パターン発見 / 機械学習 / 説明可能AI / メモリ共有型並列 / 連関分類器

Outline of Final Research Achievements

In this study, we aimed to establish discriminative pattern mining techniques as a foundation of human-centric machine learning. Specifically, we started with ECHO, a discriminative pattern mining method previously proposed by this study's principal investigator, and evolved it methodologically. We also examined the usefulness of ECHO mainly in medical domains. In this study, we achieved a certain progress in each of four sub-goals: evaluating the interpretability of ECHO-produced patterns in medical domains, development of a monitoring tool for ECHO, shared-memory parallelization of ECHO, and development of an associative classifier that uses ECHO-produced patterns. We also confirmed the importance of transparency in machine learning through an extensive survey of explainable AI.

Free Research Field

知能情報学

Academic Significance and Societal Importance of the Research Achievements

近年の人工知能の中核を成す機械学習技術ではその不透明性，近似的な解，ハイパーパラメータチューニングの必要性等，人間側が機械学習手法を確実・容易に理解・応用する上で障害となる問題点が残されている．本研究ではこれらの問題点をあらかじめ避けられる識別パターン発見手法であるECHO法に注目し，その方法論上の改良・発展および有用性の検証を行った．時間的な制約により，4つの副目標全てを完全に達成できた訳ではないが，これらの副目標の重要性が改めて確認でき，この方向で研究を進めていく意義を示すことができた．また，本研究を通じ，学術界・産業界で注目される説明可能AI分野に対する貢献ができたと考えている．