报告题目：The Dual Majorizing Measure Theorem for Canonical Processes

报告人：胡轩昂（山东大学）

报告时间：10月14号，星期三 ，上午8:00-9:30

报告地点：五教5307

摘要：

The majorizing measure theorem provides a geometric characterization of the expected supremum of a stochastic process. In this talk, we first review Talagrand's majorizing measure theorem for Gaussian processes and Latała's characterization of canonical processes with independent log-concave-tailed coordinates. We then present a dual version of Latała's theorem, which characterizes the expected supremum in terms of parameterized separation trees. The proof is based on a contraction principle for a family of intrinsic distances and a corresponding growth condition. Finally, we discuss how these methods lead to a deterministic polynomial-time approximation algorithm for finite index sets under suitable computational assumptions.

This is joint work with Vladimir V. Ulyanov and Hanchao Wang.