题目：Data-Centric AI: Taming AI-ready Feature Space—From Decision-Making to Generative-AI

主讲人：堪萨斯大学 王东杰 助理教授

主持人：统计学院 黄雁用教授

时间：2024年7月3日（周三）下午15:00-16:00

地点：西南财经大学柳林校区诚正楼统计学院会议室

报告摘要：

Unlike humans, AI systems are brittle and not robust. They often struggle when faced with novel situations, and are highly sensitive to small perturbations, which can lead to catastrophically poor performance. These systems comprise two main components: the model and the data. In recent decades, research primarily focused on models, emphasizing advanced structures or algorithms to enhance AI performance. However, the data-centric aspect consumes most of the time and resources of human experts and greatly influences AI systems. Furthermore, the gains from the model-centric part are reaching a plateau. Thus, Dongjie shifted his research focus towards data-centric AI. In this presentation, Dongjie will introduce the concept of “transformation learning”, which refines the feature space by feature selection and generation. He will show two innovative research perspectives: 1) decision-making perspective and 2) generative-AI perspective. The first perspective formulates feature selection or generation as Markov decision-making processes and employs reinforcement learning to stimulate and accelerate both tasks. The second perspective views the two tasks from the generative AI side, which embeds discrete feature learning knowledge into a continuous space and quickly pinpoints the best embedding within the learned space. Then, the best embeddings are used to reconstruct the optimal feature space. Finally, he will conclude by discussing the future research directions and visions for the data-centric AI domain.

主讲人简介：

王东杰是堪萨斯大学电气工程与计算机科学系的助理教授。他的学术研究重点包括数据中心的人工智能、因果图学习、时空数据挖掘、用户画像和图挖掘。在攻读博士学位期间，他通过在诺基亚贝尔实验室、NEC美国实验室和京东硅谷研究中心的实习，获得了宝贵的行业经验。他已在顶级期刊（如TKDE, KAIS）和会议（如NeurIPS, KDD, AAAI, WWW）上发表了30多篇论文。值得注意的是，他的三篇论文被ICDM和SIGSPATIAL被评为Best Paper Runner-up，他的NeurIPS论文被评为Spotlight。他在自动化城市规划方面的创新工作受到了关注，得到了Synced AI和UCF Today的媒体报道。除了研究贡献外，他还积极为学术界做出贡献，担任KDD, IJCAI, AAAI, WSDM, CIKM, TNNLS, KBS, TKDD等顶级会议和期刊的程序委员会成员。