报告题目：Continuous Tensor Representations：Theory, Methods, and Applications
报 告 人：赵熙乐
工作单位：电子科技大学
报告时间：2026年11月6日上午9:00 -11:30.
地点：腾讯会：446796726
报告摘要：Classical discrete tensor representations can only represent multidimensional data on meshgrid, limiting their applicability in real-world scenarios beyond meshgrid. To break this limitation, we suggest a new continuous tensor representation as an alternative to the classic discrete tensor representation, providing a unified framework for modeling both meshgrid and off-meshgrid data. Specifically, the suggested representation factorizes the underling multivariate tensor function into simple， structured factors matrices/tensors/functions with interactions between them, which reveals the intrinsic structure behind data. We establish the theoretical foundations for the suggested representation from the complementary perspectives of linear algebra and function approximation. Extensive experiments across image processing, machine learning, scientific computing, and computer graphics demonstrate the expressive power and versatility of the proposed representation compared with state-of-the-art methods.
报告人简介：赵熙乐，电子科技大学教授/博导。入选国家高层次青年人才、四川省学术和技术带头人。第一/通讯在权威期刊SIAM 系列(SIIMS和SISC)和IEEE系列(TPAMI、TIP、TSP、TKDE等)中国科学及顶会CVPR等发表研究工作。研究成果获四川省自然科学奖一等奖、四川省科技进步奖一等奖、教育部科学研究优秀成果奖二等奖、华为火花奖。主持国家自然面上项目、四川省杰青项目、中俄数学挑战基金项目、华为项目等。担任IEEE Internet of Things Journal等期刊编委。
