Speaker: Dongwoo SheenVenue: 210 Haina Complex Building 2Abstract: High dimensional problems (of dimension >> 4) occur in many fields.Typical examples include(1) String theory (10D, 11D), Kaluza-Klein theories (5D),(2) Financial Mathematics: Multi-asset options pricing (high-dimensional PDEs)(3) Parametric Studies: Physical space + parameter space dimensions(4) Uncertainty Quantification: 3D space + 1D time + N stochastic dimensions(5) Machine Learning: Feature spaces with hundreds of dimensionsSome high dimensional problems can be reduced to lower dimension, by using suitable dimension reduction techniques. However, some other challenging problems should be solved as accurate as possible without dimension reduction.In this talk, we review several dimension reduction strategies briefly. Then we move to discuss several issues in essentially high dimensional problems, with the aim of accurate computing. In order to challenge the curse of dimensionality, sometimes we need to break our common senses, looking at problems from a completely different directions. Some examples from the finite element methods will be given.

Speaker: Shan ZhaVenue: Meeting Room 705, School of Medical Complex Building, Zijingang CampusAbstract: Professor Shan Zha is the James A. Wolff Professor of Pediatrics at Columbia University Medical Center, an elected member of the American Society for Clinical Investigation (ASCI), and an internationally renowned scholar in the interdisciplinary field of DNA damage repair and immunology. Her research has long focused on DNA double-strand breaks and genomic instability. Using transgenic mouse models and cell biology techniques, she has systematically elucidated the central roles of key repair factors such as ATM, DNA-PKcs, and PARP in maintaining genomic stability, regulating immune system development, and suppressing lymphomagenesis. In recent years, she pioneered the discovery of novel roles for the DNA repair proteins Ku and DNA-PKcs in RNA metabolism and innate immune regulation, opening new directions in the intersection of genome integrity and immune response.Professor Zha has an extensive publication record, with over 70 papers in top international journals such as Nature, Molecular Cell, and PNAS. As an independent principal investigator, she has led eight NIH R01 grants and one NIH P01 multidisciplinary program project. She has long served on the editorial boards of journals including Nucleic Acids Research and JCB, and is a reviewer for leading journals such as Nature and Science. In 2020, she launched and organized the international online seminar series “SocialDNAling,” bringing together scholars from around the world. She has also served as chair multiple times for Gordon Research Conferences (GRC) and Cold Spring Harbor Laboratory (CSHL) meetings, making significant contributions to advancing the field and promoting international collaboration.

Speaker: Sami UllahVenue: Meeting Room 123, College of Life Sciences, Zijingang CampusAbstract: Land ecosystems including forests absorb ~25% of the total CO2 emissions from anthropogenic sources. This sink is predicted to increase under increasing CO2 concentration. However, carbon capture in forests is also controlled by nutrient availability including nitrogen and these interactions are poorly understood to support and validate when modelling the role of forests in climate change mitigation.To elucidate the feedbacks between elevated CO2, C capture and nutrient availability, the University of Birmingham Institute of Forest Research (BIFoR) established a Free-Air CO2 Enrichment (FACE) facility exposing patches of temperate forests to +150 ppm CO2 above the ambient since 2017. In this seminar, the carbon capture and nitrogen cycling responses to nine year of CO2 enrichment will be highlighted to show the tight coupling of nitrogen cycling with carbon capture in mature temperate forests under future climates.

Speaker: Peter JimackVenue: Lecture Hall 210, Haina Complex Building 3, Zijingang CampusAbstract: Recent developments in artificial intelligence (AI) algorithms and hardware are having a major impact on Computational Science. The traditional paradigm in Scientific Computing (SC), typically based upon mathematical models of the underlying phenomena followed by discrete approximation and a numerical solution, are being complemented and challenged by machine learning (ML) and other AI capabilities. In this presentation, I will consider a range of examples in which we have used ML techniques to enhance the traditional SC workflow, taking mesh generation and preconditioning as examples, followed by a discussion of ways in which physics-aware ML techniques may be applied in place of numerical schemes in applications such as fluid dynamics and nonlinear elasticity.

Speaker: Steven LaureysVenue: Meeting Room 313, Haina Complex Building 3, Zijingang CampusAbstract: How does the brain generate conscious experience, and how can it recover when severely injured? Advances in neuroimaging reveal the dynamic brain networks underlying awareness, offering new insights into both pathology and human potential.In this keynote, Prof. Steven Laureys will synthesize three decades of research on disorders of consciousness, including coma, unresponsive wakefulness syndrome, minimally conscious state and locked-in syndrome. Multimodal imaging shows how large-scale brain networks supporting sensory awareness and self-related processing can be disrupted by injury, yet retain remarkable capacity for recovery through neuroplasticity.Beyond clinical neurology, altered states such as dreaming, psychedelic states and meditation provide complementary windows into the mechanisms shaping conscious experience. Studies of long-term meditators show how structured mental training can induce measurable changes in brain connectivity related to attention, emotion regulation and cognitive flexibility.

Speaker: SHEN ZhongweiVenue: Lecture Hall 210, Haina Complex Building 2, Zijingang CampusAbstract: The resolvent estimates for the Stokes operator play an essential role in the functional analytic approach of Fujita and Kato to the Navier-Stokes equations. This talk is concerned with the study of resolvent estimates and the analyticity of the semigroup in LP for the Stokes operator in domains with rough boundaries. In the case of smooth domains (C2), the resolvent estimates are known to hold for all 1< p ≤ ∞. If the domain is Lipschitz, the estimates were established or a limited range of p,depending on the dimension. In this talk, I will present some recent work, joint with Jun Geng, or the case of C1 domains. In particular, I will discuss a key step in the case p=∞, which involves some new estimates that connect the pressure to the gradient of the velocity in the Lq average, but only on scales above certain level. 

Speaker: CHENG MengjieVenue: A523 School of Management, Zijingang CampusAbstract: We study how media firms can use LLMs to generate news content that aligns with multiple objectives—making content more engaging while maintaining a preferred level of polarization/slant consistent with the firm’s editorial policy. Using news articles from The New York Times, we first show that more engaging human-written content tends to be more polarizing. Further, naively employing LLMs (with prompts or standard Direct Preference Optimization approaches) to generate more engaging content can also increase polarization. This has an important managerial and policy implication: using LLMs without building in controls for limiting slant can exacerbate news media polarization. We present a constructive solution to this problem based on the Multi-Objective Direct Preference Optimization (MCDPO) algorithm, a novel approach that integrates Direct Preference Optimization with multi-objective optimization techniques. We build on open-source LLMs and develop a new language model that simultaneously makes content more engaging while maintaining a preferred editorial stance. Our model achieves this by modifying content characteristics strongly associated with polarization, but that have a relatively smaller impact on engagement. Our approach and findings apply to other settings where firms seek to use LLMs for content creation to achieve multiple objectives, e.g., advertising and social media.

Speaker: Prof. Luca ValentiVenue: E122, College of Animal Sciences, Zijingang CampusAbstract: Professor Luca Valenti is the head of Biological Resource Center, the Precision Medicine Lab and the Omics Lab of Fondazione IRCCS Ca' Granda Ospedale Policlinico Milano, University of Milan as an internist and clinical hepatologist at the Department of Transfusion Medicine. He has authored more than 200 publications as main author in journals, including The New England Journal of Medicine, Nature Genetics, Gut, and Journal of Hepatology (Scholar: 50,000 citations, H-index 108).His main research interests are metabolic and genetic liver diseases, with a special focus on steatotic/fatty liver disease (SLD) and iron metabolism disorders. He conducts genetic studies in well-characterized clinical and population-based cohorts, integrated with translational research with evaluation of biomarkers and biological pathways in clinical samples, and experimental studies in vivo and in vitro multilineage 3D models.

Speaker: Ni YiVenue: 210 Haina Complex Building 2, Zijingang CampusAbstract: Knot Floer homology is an invariant of knots in closed 3-manifolds. This invariant captures a great deal of information about the topology of knot complements. In particular, when a knot is fibered, the information contained in knot Floer homology provides insight into the monodromy of the fibration. I will survey recent work on this topic, including results on bounding the number of fixed points and the characterization of right-veering monodromy.