2023-01-012024-05-13https://scholars.lib.ntu.edu.tw/handle/123456789/651714子計畫一「可信賴的智慧學習與人本計算」開發前瞻智慧學習與人本資料分析技術，提升資料分析之實用性、可解釋性及可依賴性。計畫開發「可信賴的人本人工智慧技術」及將可信賴的人工智慧技術用於「需要高度使用者隱私保護及模型可靠性的人本應用」。在可信賴的人工智慧技術方面，預計研究防範AI模型開發者得知使用者的隱私資訊、防範使用者對AI模型的惡意攻擊、防範AI開發者提供惡意的演算法或模型、人本機器學習技術。而在人本應用方面，將開發智慧助理(推測使用者隱性意圖、引導其回憶等)及預防醫療(疾病檢測等)之應用。主要研究關鍵議題為人工智慧模型偏見和弱點自動檢測與修復。 Subproject 1, "Trustworthy Intelligent Learning Techniques and Human-Centered Computation," develops forward-looking intelligent learning and human-centric data analysis technology to improve the practicability, interpretability, and reliability of data analysis. The project aims to develop `trustworthy human-centered artificial intelligence technology` and apply trusted AI technology to `human-centered applications that require high user privacy protection and model reliability.` Regarding trustworthy AI technology, the research aims to prevent AI model developers from accessing users` private information, safeguard users against malicious attacks on AI models, prevent malicious algorithms or models provided by AI developers, and focus on human-centric machine learning techniques. As for human-centered applications, the project will develop intelligent assistants (predicting users` implicit intentions, guiding memory, etc.) and preventive healthcare applications (disease detection, etc.). The main research issue is the Automated Detection and Repair of Vulnerabilities and Bias in Artificial Intelligence Models.智慧學習;人本計算;可信賴的人本人工智慧技術;人工智慧模型偏見;弱點自動檢測與修復;Intelligent Learning Techniques; Human-Centered Computation; Trustworthy Human-centered Artificial Intelligence Technology; Bias in Artificial Intelligence Models; Automated Detection and Repair of Vulnerabilities