Principled Learning Models: Leveraging Belief Change Theory for Transparent and Reliable Classifiers

• Project/Area Number: 25K00375
• 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: National Institute of Advanced Industrial Science and Technology
• Principal Investigator: Nicolas Schwind 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (60646397)
• Project Period (FY): 2025-04-01 – 2028-03-31
• Project Status: Granted (Fiscal Year 2025)
• Budget Amount: ¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000); Fiscal Year 2027: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000); Fiscal Year 2026: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000); Fiscal Year 2025: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
• Keywords: Principled AI / Belief Change Theory / Knowledge Representation / Symbolic Learning / Model Interpretability

Outline of Research at the Start

This research project aims to create a new architecture for classification learning models based on principles from Belief Change Theory. It addresses the limitations of current AI models, which often operate as black boxes with no transparent decision-making process. The project includes establishing formal parallels between Belief Change Theory and classifier learning, adapting rational principles, designing specific algorithms, and empirically comparing the new models with existing methods. The goal is a principled, transparent, and reliable learning model architecture.