Scholarlink

<p>The explainability of models has emerged as a critical concern across academia and industry. In this paper, we discuss the definition of explainability and how to integrate domain-specific prior knowledge to enhance transparency in Artificial Intelligence (AI) systems. Additionally, we explore the different approaches to achieve explainability in graph computing including local versus global explanations, post hoc explanations, and explainability by design. Then, we conduct several effective strategies for embedding prior knowledge into machine learning models aiming at enhancing the logical coherence and human interpretability of models. Finally, we highlight how graph computing techniques enhance explainability by capturing complex relational structures, supporting causal inference, and providing intuitive visualization methods. This paper offers a comprehensive perspective to better understand and address the challenges of explainability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable Modeling Based on Prior Knowledge Embedding and Higher-Order Correlations

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Dong Zhang,
  • Shuai-Chen Zhuo,
  • Yuan Sun,
  • Xi-Jing Wang,
  • Hong-Cheng Han,
  • Mei-Qin Liu,
  • Juan Wang,
  • Jue Jiang

摘要

The explainability of models has emerged as a critical concern across academia and industry. In this paper, we discuss the definition of explainability and how to integrate domain-specific prior knowledge to enhance transparency in Artificial Intelligence (AI) systems. Additionally, we explore the different approaches to achieve explainability in graph computing including local versus global explanations, post hoc explanations, and explainability by design. Then, we conduct several effective strategies for embedding prior knowledge into machine learning models aiming at enhancing the logical coherence and human interpretability of models. Finally, we highlight how graph computing techniques enhance explainability by capturing complex relational structures, supporting causal inference, and providing intuitive visualization methods. This paper offers a comprehensive perspective to better understand and address the challenges of explainability.