Speaker: SHI ZhentaoVenue: Building 2, room 206, Haina yuan, Zijingang CampusAbstract: When the number of assets is larger than the sample size, the minimum variance portfolio interpolates the training data, delivering pathological zero in-sample variance. We show that if the weights of the zero variance portfolio are learned by a novel “Ridgelet” estimator, in a new test data this portfolio enjoys out-of-sample generalizability. It exhibits the double descent phenomenon and can achieve optimal risk in the overparametrized regime when the number of assets dominates the sample size. In contrast, a “Ridgeless” estimator which invokes the pseudoinverse fails in-sample interpolation and diverges away from out-of-sample optimality. Extensive simulations and empirical studies demonstrate that the Ridgelet method performs competitively in high-dimensional portfolio optimization.

Speaker: Mingchao CaiVenue: Building 2, room 204, Haina yuan, Zijingang CampusAbstract: Thermo-poroelastic models capture the interplay between elastic porous material deformation, fluid flow, and thermal effects under non-isothermal conditions. This talk presents a four-field formulation for the linear thermo-poroelastic model and introduces two novel algorithms. The first focuses on constructing parameter-robust preconditioners for the resulting linear system, proposing two approaches: one reorganizes variables into a 2-by-2 block structure, while the other directly addresses the 4-by-4 coupled operator. Both preconditioners exhibit robustness to parameter variations and mesh refinement. The second algorithm is a decoupled iterative finite element method, for which stability and optimal convergence are rigorously proven. Numerical experiments are provided to validate the effectiveness and efficiency of the proposed methods.

Speaker: HUANG XuehaiVenue: Room 101, building 2, Haina yuanAbstract: In this talk, we begin with the development of distributional finite element complexes, including the distributional div div and curl div complexes. In the main part of the talk, we discuss applications of these complexes to partial differential equations, such as the biharmonic equation, fourth-order elliptic singular perturbation problems, quad-curl problems, and the Stokes equation. We also present a distributional formulation and a corresponding discretization for strain gradient elasticity.

Speaker: YU BangxingChair: QIU YangVenue: Room A215, School of Management, Zijingang campusAbstract: This study examines how improved judicial independence affects corporate governance using the staggered establishment of circuit courts in China. By reducing local protectionism, these courts increase firms’ exposure to litigation risk. We find that independent director compensation rises following the reform. The evidence supports an efficient contracting explanation rather than rent extraction. The increase is stronger in regions with greater prior protectionism and for directors located farther from the firm. The effect persists in firms with strong external monitoring and high product market competition, where agency concerns are less pronounced. Additional analyses show that the reform leads to more active board oversight, reflected in more dissenting votes and greater director expertise. Overall, our study suggests that improvements in judicial independence reshape corporate contracting by increasing the value of effective board monitoring in response to higher legal risk

Speaker: Andreas ChristmannVenue: Building 2, room 204, Haina yuan, Zijingang CampusAbstract: The bootstrap is a computer-based resampling method that can provide good approximations to the finite sample distribution of a given statistic. In this talk some methods to use the empirical bootstrap approach for stochastic gradient descent (SGD) to minimize the empirical risk over a Hilbert space are investigated from the view point of algorithmic stability and statistical robustness. Two types of approaches are based on averages and are investigated from a theoretical point of view. Another type of bootstrap SGD is proposed to demonstrate that it is possible to construct purely distribution-free pointwise confidence intervals and distribution-free pointwise tolerance intervals of the conditional median function using bootstrap SGD.

Speaker: WU ShukunVenue: Room 204, building 2, Haina yuan, Zijingang campusAbstract: The local smoothing conjecture for wave equations was raised by Sogge, initially aiming to understand Stein's spherical maximal function. Because of its close connection to the Fourier transform of the surface measure of the sphere, the local smoothing conjecture has become a central topic in harmonic analysis. In this talk, I will discuss some quantitative smoothing estimates in both R^n and compact Riemannian manifolds. I will compare wave equations with (linear) Schrodinger equations and explain why the quantitative smoothing problem for wave equations is more challenging.

Speaker: Arnak DalalyanVenue: Room 1417, Administration Building, Zijingang CampusAbstract: Generative modeling aims to produce new random examples from an unknown target distribution, given access to a finite collection of examples. Among the leading approaches, denoising diffusion probabilistic models (DDPMs) construct such examples by mapping a Brownian motion via a diffusion process driven by an estimated score function. In this work, we first provide empirical evidence that DDPMs are robust to constant-variance noise in the score evaluations. We then establish finite-sample guarantees in Wasserstein-2 distance that exhibit two key features: (i) they characterize and quantify the robustness of DDPMs to noisy score estimates, and (ii) they achieve faster convergence rates than previously known results. Furthermore, we observe that the obtained rates match those known in the Gaussian case, implying their optimality. (Joint work with V. Arsenyan and E. Vardanyan)

Speaker: Yafei LiangVenue: Room 426, School of Economics, Zijingang CampusAbstract: This study examines how large language models (LLMs) affect language barriers and disparities in publication outcomes and research impact for English-as-a-Foreign-Language (EFL) scholars. Using the release of ChatGPT-3.5 in November 2022 as a natural experiment, we analyze paper metadata from the Social Science Research Network (SSRN) across seven subject areas. We find that LLM usage increased significantly after November 2022 and that EFL-authored papers show greater improvements in language proficiency than non-EFL papers, particularly in Social Science. However, this improvement in writing quality is accompanied by only a modest increase in publication success and no detectable increase in research impact. These findings highlight both the promise and limitations of LLMs: while language gaps narrow, deeper disparities in scientific impact persist.

Speaker: Kallol RayVenue: Room 203, Haina Building 2, Zijingang CampusAbstract: This work investigates both steady and unsteady nonlinear systems that couple the generalized convective Brinkman-Forchheimer model with a system of advection-diffusion equations, commonly referred to as double-diffusion equations. The existence and uniqueness of weak solutions to the governing equations are established using Galerkin’s method. Subsequently, H(div)-conforming discontinuous Galerkin (DG) discretizations are formulated for the considered models, yielding exactly divergence-free velocity approximations. For the unsteady model, a second-order semi-implicit backward differentiation formula (BDF2) scheme is employed for temporal discretization. A rigorous analysis is then carried out to establish the well-posedness of the discrete problems. Optimal a priori error estimates are derived, ensuring that the velocity errors are pressure-robust. Furthermore, when the diffusion coefficients are constant, the velocity error estimates are Re-semi-robust at high Reynolds numbers. Numerical experiments are presented to corroborate the theoretical results and to demonstrate the performance of the proposed methods.