应威廉希尔中文官方网站的邀请，加拿大埃德蒙顿阿尔伯塔大学Witold Pedrycz教授于2025年4月28日下午3点，在中关村校区主楼317会议室带来了一场以《Data-Knowledge Environment and Knowledge Landmarks in Machine Learning》为题的学术报告。报告会由郭思尼老师主持，学院众多师生参加了本次报告会。
Witold Pedrycz 教授在报告中围绕机器学习中数据与知识的关系展开深入探讨。他提到，机器学习近年来凭借对海量数据的有效运用取得显著进展，大语言模型和基础模型的成功便是有力例证。但当前机器学习高度依赖数据，知识的作用却常被忽视。为此，在本次报告中，教授大力倡导将知识与数据进行统一设计的机器学习（KD-ML）这一全新范式。
Witold Pedrycz教授阐释了KD环境的基本原理，回顾其发展历程并点明关键要点。他深入探讨了面向问题的知识起源、知识分类及其主要特征。数据多为数值形式，而知识在符号层面进行形式化和表示，二者抽象层次差异巨大，如何协调成为挑战。在此过程中，信息粒发挥着关键作用。教授还对知识进行分类，区分了科学知识和常识知识，并阐述了相应的知识表示方案。围绕知识导向的机器学习设计，教授讨论了多个主要类别。以物理学为基础的机器学习借助科学知识，通过知识导向的约束扩充数据驱动模型，对数据驱动模型进行粒度扩展，以及基于规则传递的知识构建机器学习模型。这些方法有效避免了数据盲目性带来的不利影响。此外，教授还介绍了KD统一环境的相关方案和学习方法，包括基于大语言模型的知识获取研究。
报告中，教授引入知识地标的概念，它是基于历史数据中的实验证据总结而成，通过无监督学习和聚类方法构建，在提升抽象层次的同时，实现对数据的有效概括。教授还介绍了基于高斯过程回归的知识地标操作框架，能够为模型提供概率信息，增强模型的预测能力。在规则基模型设计方面，教授探讨了基于名义参数值和参数空间采样的设计方法，并通过污染物泄漏案例研究，展示了模型在实际应用中的效果。
报告结束后，Witold Pedrycz教授与现场师生展开热烈讨论。本次报告为师生们带来机器学习领域的前沿理念和方法，得到师生们的一致赞誉。
汇报人简介：
Professor Witold Pedrycz is a distinguished academic at the University of Alberta, Edmonton, Canada, renowned for his contributions to Computational Intelligence, Granular Computing, and Machine Learning. He has extensive expertise in areas such as data mining, knowledge discovery, and fuzzy logic. His research focuses on developing innovative methodologies that integrate data and knowledge in machine learning, aiming to enhance the efficiency and effectiveness of computational models.
Professor Pedrycz is an IEEE Life Fellow and holds positions at both the University of Alberta and the Systems Research Institute of the Polish Academy of Sciences. He is a foreign member of the Polish Academy of Sciences and a Fellow of the Royal Society of Canada. He has received numerous prestigious awards, including the Norbert Wiener Award from the IEEE Systems, Man, and Cybernetics Society, the IEEE Canada Computer Engineering Medal, and the Cajastur Prize for Soft Computing.
He has made significant advancements in the field through his research on the unification of data and knowledge in machine learning. His work includes the development of the Knowledge-Data Machine Learning (KD-ML) paradigm, which emphasizes the coordinated engagement of data and knowledge in model design. He has also explored the role of knowledge landmarks in machine learning, providing new perspectives on model abstraction and generalization. Professor Pedrycz serves as the Editor-in-Chief of WIREs Data Mining and Knowledge Discovery (Wiley) and Co-editor-in-Chief of J. of Data Information and Management (Springer), playing a key role in promoting academic exchanges and research progress in related fields.
