Discovering New Knowledge by Combining Symbolic Logic and Deep Learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K21302
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Multi-year Fund
• Review Section: 1002:Human informatics, applied informatics and related fields
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: Phua Yin Jun 東京工業大学, 情報理工学院, 助教 (20963747)
• Project Period (FY): 2022-08-31 – 2024-03-31
• Keywords: 深層学習 / ニューロシンボリック / 記号推論 / 知識発見

Outline of Final Research Achievements

This study focused on methods for discovering new human-readable knowledge by utilizing deep learning methods. First, we proposed a method that utilizes ensemble, where multiple models are trained, to stabilize the results for extracting features that are considered important by the models. Next, by utilizing a method to generate training data, which has been proposed for symbolic knowledge extraction methods, is applied to using generative AI in the incremental learning field. With this, we show that methods proposed for symbolic knowledge extraction can also be generalized to deep learning methods. Furthermore, we also proposed graph neural network models that are robust to noise that are inherent in graphs.

Free Research Field

知能情報

Academic Significance and Societal Importance of the Research Achievements

学術的意義として，実験データから重要な要素を抽出する基盤技術に貢献することができた．また，記号推論でよく使われる手法として，学習データを生成する手法は深層機械学習にも応用が可能であることを示した．さらに，生成AIの学術的応用を示すこともできた．社会的には，ノイズに対して頑健な手法を提案することで，実世界のデータをそのまま応用することが可能となる技術の開発に貢献した．本研究で開発した手法をさらに展開させることにより，実験データから新たな知識を抽出することができるAI技術へつながることが考えられる．