教师发展中心“学者论坛”活动特别邀请厦门大学钟威教授来校作学术交流。具体安排如下,欢迎广大师生参加。
一、主 题:SIRI——Tensor Rank Selection via Stability Index with Random Initializations
二、时 间:2026年11月13日(星期五)14:00-15:30
三、地 点:清水河校区六号科研楼A344
四、主讲专家:厦门大学钟威教授
五、主持人:数学科学学院徐立伟教授
六、内容简介:
Selecting the rank of high-order tensors is a fundamental yet ill-posed problem, particularly in high-dimensional and low signal-to-noise settings. We propose the stability index with random initializations (SIRI), a framework that leverages the sensitivity of non-convex tensor optimization to random initializations, typically treated as a computational nuisance, as a statistical signal for model selection. SIRI quantifies the reproducibility of tensor factors across independent runs via a novel Hadamard geometric aggregation of pairwise correlations. Theoretically, SIRI is determined by the underlying geometry of the parameter space, leading to a phase transition. At the true rank, the index converges to unity at a quadratic rate in the noise level, in contrast to the linear convergence of factor estimates. Under over-parameterization, stability strictly decays via two distinct mechanisms: redundant factors may share the true signal and drift along a continuous solution manifold, bounding stability away from unity; alternatively, they fit high-dimensional noise and become asymptotically orthogonal, inducing a geometric decay law approaching the ratio of the true to candidate rank ($R_0/r$). Simulations and applications to fluorescence spectroscopy and tensor regression demonstrate that SIRI provides a statistically interpretable and theoretically grounded criterion for discovering latent low-dimensional structures.
七、主讲人简介:
钟威,厦门大学王亚南经济研究院、经济学院统计学与数据科学系,南强特聘教授(二级教授)、博士生导师。国家级领军人才,国家自科优秀青年基金获得者(2019)。同时兼任厦门大学国际合作与交流处处长、台港澳事务办公室主任、海外办学事务办公室主任。主要从事高维数据统计分析、统计学习算法、计量经济学等研究。先后担任JASA、ISR等6个国内外期刊的编委(AE),在AOS、JASA、Biometrika、JoE、JBES、JMLR、AOAS、《中国科学-数学》、《统计研究》等国内外统计学权威期刊发表(含接收)50多篇论文。曾获霍英东教育基金会高等院校青年科学奖、第九届高等学校科学研究优秀成果奖、全国宝钢优秀教师奖,以及厦门大学青年教师技能比赛特等奖,厦门大学“我最喜爱的十位老师”,厦门大学南强卓越教学团队负责人,主编教育部统计学101计划教材《人工智能的统计基础》。
八、主办单位:教师发展中心
承办单位:数学科学学院计算机科学与工程学院(网络空间安全学院)