安徽省非线性科学学会第五次会员大会

暨2026年学术年会会议日程

一、会议时间：2026年9月27日

二、会议地点：中国科学技术大学东区第五教学楼            

三、会议日程：

（一）9月26日：会议报到。

时间：12:00-19:00

地点：专家楼或第五教学楼                  

（二）9月27日：会员大会及年会。

（1）8:30-8:40，会议开幕式                                

（2）8:40-9:20，大会报告：曾长淦（40分钟）

（3）9:20-10:00，大会报告：牛 谦（40分钟）

（4）10:00-10:20,茶歇，大会合影

5. 10:20-11:00，大会报告：范洪义（40分钟）        

 （6）11:00:11:20，大会报告： 徐锦峰（20分钟）

（7）11:20-11:40，大会报告： 宋卫国（20分钟）                                              

（8）11:40-12:00，大会报告： 高  见（20分钟）            

 （9）12:00-13:00，午 餐    

  （10）13:15-14:45会员大会（略）              

（11）14:45-15:00，茶歇        

（12）15:00-16:00，分组会议报告

量子多体问题、AI for Science and Medicine、流体力学、生物统计和医疗大数据、复杂网络和非线性动力学、交通流和智能交通系统、安全科学与工程                                                                

（13）16:00-16:15，茶歇      

（14）16:15-17:15，分组会议报告

（15）17:15-17:30，茶歇      

（16）17:30-18:00，大会分组会议总结、闭幕式

安徽省非线性科学学会第五次会员大会暨学术年会

会议通知

尊敬的各位老师：

为了加强安徽省非线性科学研究领域同行的学术交流与合作，扩大安徽省非线性科学研究的影响，安徽省非线性科学学会第五次会员大会暨学术年会将于2026年9月27日在安徽省合肥市中国科学技术大学举行。安徽省非线性科学学会换届会议也将同时举行。

我们真诚欢迎您及您的课题研究团队成员与会交流，期待分享您的研究成果，同时欢迎您将会议通知转发给感兴趣的会员及同事。请在2026年9月22日前扫描下方会议二维码注册，以便会务组进行进一步安排。

会议注册二维码

本次会议将邀请安徽省非线性科学学会的多位理事及其他著名专家做大会报告，为所有参会的老师和同学带来非线性科学研究领域最新的学科前沿分享及研究成果的交流与探讨。

本次会议建议的学术交流范围包括非线性科学的以下研究方向（但不仅限于这些方面）：

量子多体问题的理论及应用、流体力学、AI for Science and Medicine、复杂网络和非线性动力学、交通流和智能交通系统、安全科学与工程等。

本次会议收取注册费，会议期间的餐费由主办方承担，交通费和住宿费敬请自理。期待您参加本次会议。非常感谢您对本次会议的大力支持！

一、会议相关安排如下：

报到时间：9月26日，15:30-18:00

报到地点：科大东区 专家楼一楼大厅

会议时间：9月27日全天

会议地点：科大东区 第五教学楼            

就餐地点：科大东区 专家楼二楼自助餐厅

（报到时，如已网上转账缴纳注册费请出示凭证办理报到手续。）

二、注册费及缴费方式：

本次会议注册费由合肥天青会议会展服务有限公司代为收取。

1. 注册费缴纳方式有三种：（1）网上转账缴费；（2）扫描付款码缴费；（3)现场缴费。网上转账缴费请注明姓名、本单位的名称和纳税人识别号、邮箱地址等信息。网上转账信息如下：

名称：合肥天青会议会展服务有限公司

账号：187215013422

开户行：中国银行股份有限公司合肥蜀山支行

会议注册费发票由合肥天青会议会展服务有限公司开具。

（温馨建议：各位参会代表可提前网上转账缴费或扫描下方注册费付款码缴费，开会期间可及时收到发票。）

2、注册费标准：

所有参会人员注册费标准均为600元/人。

3. 注册费付款码

3. 会务组联系人：

                               陈梓韬chenzitao@ustc.edu.cn      电话：13030222005

                                                                沈俊尧sjy2003@mail.ustc.edu.cn  电话：18205006799

                                                                唐  慧tanghui@mail.ustc.edu.cn    电话：18225796951

                                                    安徽省非线性科学学会

                                                       2026年9月18日

安徽省非线性科学学会
第五次会员大会报告汇总（部分）

大会场

IWOP方法对泛函分析和对无穷维矩阵线性代数的发展之影响

范洪义　中国科学技术大学

摘要：为了发展量子力学的狄拉克符号法，报告人发明了IWOP方法，讨论它与泛函分析与无穷维线性代数积分的关系

医学数据分析中的两个应用：时频分析与多模态融合

Jinfeng Xu，香港城市大学

摘要：本报告结合近期与合作者完成的两项工作，讨论医学数据分析中若干具有共性的问题，以及我们在处理这些问题时采用的方法。第一项工作关注生理信号的时频表征。医学信号通常是非平稳的，具有诊断价值的特征往往较为微弱且集中于低频段，现有时频分析方法在此类情形下的分辨能力受到限制。我们对变换本身加以改造，引入随频率自适应的窗函数与时间方向的重排，以获得更为精细的时频表征，并在此基础上进行分类。第二项工作关注临床数据的缺失与多模态信息的整合。在实际数据中，检验指标的缺失比例往往很高，各模态的可得性亦不一致，常规插补方法难以保持变量之间的相关结构。我们采用生成式模型对缺失指标按分布进行补全，并在个体层面将影像信息与检验信息加以融合。

行人与疏散动力学的多类型实验与快速建模研究

宋卫国　中国科学技术大学

摘要：国内外大型公众聚集场所踩踏事故时有发生，造成重大人员伤亡。行人疏散动力学是评估人群风险、优化建筑疏散设计的重要基础，然而目前该领域仍存在基础数据不足、模型验证不充分的现实问题，不同人群构成、人流交汇条件下的运动规律，还需要大量研究，计算效率也需要不断提升。为此我们设计开展了系列大规模人群疏散试验，并建立了快速计算方法。一方面聚焦高密度人群场景，针对汇流、四方向交叉流等易发生危险的工况，研究行人自组织分层等典型行为，分析速度 密度演化规律；测量和分析高密度拥挤状态下人体承受的挤压力，探究拥挤受力的变化特征，基于人体弹性压缩假设建立了挤压力分段力学模型。另一方面，针对现实场景普遍存在的异质人群，对比青年人、老年人以及老少混合群体的运动差异，研究老年个体对整体通行效率、拥堵停顿行为带来的影响。最后，基于AI算法构建快速高效的人员疏散计算方法，相比于传统方法计算效率显著提升。这些结果可望为大型活动组织、公共建筑疏散设计、踩踏风险研判提供科学技术支撑。

非平衡系统中的斑图及调控

高见　安庆师范大学

摘要：本报告围绕非平衡反应扩散系统中的自组织斑图开展结构发现与调控机制研究，主要工作沿螺旋波与图灵斑图两条主线展开。在螺旋波方向，首先通过引入周期力在多个经典振荡系统中揭示了一种无缺陷线的新型周期2螺旋波，其时空特征显著区别于传统周期2螺旋波；继而在复金兹堡-朗道方程（CGLE）中发现螺旋波存在基态与激发态两种稳态，弱脉冲仅引起波尖轨道的微小偏移，而超过阈值的强脉冲可触发两态之间的可逆跳转，由此建立了基于脉冲幅度的螺旋波态转换规则；进一步在离散时间落叶林昆虫种群模型中发现了由中心不动点附近流形与不变环竞争所诱导的复合螺旋波及超结构，并利用内部白噪声对其进行调控，刻画了特殊区域曲率与膨胀速度随噪声强度的演化规律。在图灵斑图方向，针对离散时间比率依赖捕食者-食饵模型，证明类图灵斑图的宏观结构不依赖于分岔参数，而由初始条件中高稳态区域的占比线性决定，从而将斑图结构的控制问题归结为初值调控；在此基础上，借鉴热力学自由能框架（F = E − TS），揭示噪声驱动下图灵与类图灵斑图可越过能量势垒、向自由能极小态演化的热力学稳定性机制，并识别出相应的噪声阈值。最后，报告展望了向集群自驱动粒子体系推广螺旋波等效场描述、以及提炼决定斑图结构之跨系统普适机制的研究方向。

大数据分会场

Dynamic Survival Prediction from Recurrent Adverse Events: A Semiparametric Model with Grade-Specific Nonlinear Effects

丁浩伦

摘要：Cancer drug treatment is often accompanied by adverse events that can recur during clinical trials. Their severity, frequency, and time of occurrence provide longitudinal information that may help predict subsequent survival. The prognostic relevance of an event can depend on its severity, how recently it occurred, and the event history. These relationships may be nonlinear, with different prognostic implications across severity grades. This motivates incorporating adverse event information into survival prediction.

We propose a framework for dynamic survival prediction based on recurrent adverse event histories. Event histories are summarized separately by severity grade, with the contribution of past events decreasing over time. Flexible functions capture potentially nonlinear relationships between these burdens and mortality risk. We develop estimation procedures and construct confidence intervals for the grade-specific risk functions. Simulation studies examine estimation and prediction performance, and an application to rash adverse events in 842 patients with colorectal cancer illustrates distinct prognostic patterns across severity grades.  The framework provides an interpretable basis for incorporating adverse event histories into individualized survival prediction.

Identifying Groups with Improvement Paths

董枘朋

摘要：Clustering is an important tool for grouping production units and designing group-specific management plans to improve production efficiency. Existing methods mainly group units according to their current states or reference sets which can provide guidance for reaching the efficient state. However, since organizational growth is typically gradual, these methods may overlook differences in how units progress from their current states to efficient states. We represent each such process as an improvement path consisting of a sequence of reference sets and seek to identify groups of units with similar paths.

This problem leads to two challenges: how to construct each improvement path without enumerating all possible reference sets and how to cluster paths that contain different numbers of reference sets. To address these issues, we propose an optimization-based clustering framework that combines a sequential algorithm for constructing improvement paths with a Transformer-based encoder that maps these paths to fixed-dimensional representations for clustering. We prove that the sequential algorithm has polynomial worst-case iteration complexity, and computational experiments show that it is faster than naive enumeration. Numerical simulations demonstrate that the learned representations preserve the class information contained in improvement paths and yield stable clustering results. Finally, an application to grouping industrial parks demonstrates that, compared to existing clustering methods, the proposed framework achieves higher consistency in improvement decisions within groups under all considered improvement budgets.

Efficient Estimation of Average Causal Effect with Mismeasured Exposures

随子健

摘要：Exposure mismeasurement is commonly seen in causal inference based on observational data. Recent evidence suggests that 35–61% of medical studies using diagnostic tools report nonnegligible exposure mismeasurement. For example, in the Global Enteric Multicenter Study, ELISAbased rotavirus detection showed substantial discordance with PCR-based TaqMan Array Card (TAC) testing: 20.1% of ELISA-negative cases tested TAC-positive, whereas 19.5% of ELISApositive cases tested TAC-negative. The existing methods handling exposure measurement error usually require that the mismeasurement distribution is known or can be estimated from a subset of study subjects with validated exposure statuses. This condition, however, is often violated when the mismeasurement mechanism is complex. In this paper, we develop a novel method for estimating average causal effect (ACE) using samples with exposures subject to misclassification and limited validation subsamples with correctly ascertained exposures. Our method does not require making any modeling assumptions about exposure mismeasurement. We establish some large-sample properties for the proposed method, based on which statistical inference can be easily carried out. The superiority of our proposed estimator is illustrated through extensive simulation studies. Our proposed estimator is shown to achieve a 39–53% reduction in variance over the estimator relying solely on validation data in our simulation situations. In the application to the Global Enteric Multicenter Study, the proposed method reveals a substantial adjustment of pathogen-specific mortality associations that were previously distorted by inaccurate diagnoses.

Multi-task Dynamic Pricing under Sparse Heterogeneity with High-Dimensional Data

张杰

摘要：In this paper, we propose a multi-task dynamic pricing framework for integrative analysis using data from multiple sources, enabling dynamic pricing based on the prevailing demand model, which is defined as the model supported by the majority of sites. The demand model is assumed to be a high-dimensional linear model under sparse heterogeneity where the source/taskassociated parameters are composed of a global (shared) parameter and a sparse taskspecific term. We propose the multi-task dynamic pricing (MTDP) algorithm, which leverages the shared parameter across all tasks and then refines the sparse task-specific parameters for each individual task. Under mild conditions, we derive a new highprobability error bound for the multi-task estimator and establish an upper bound for the regret of the MTDP algorithm. Our theoretical analysis demonstrates that the proposed MTDP policy outperforms the single-task pricing policy.

Monitoring Deployed Risk-Prediction Models with Test Martingales: Addressing the Challenge of Performativity  

郑潇逸

摘要：When machine learning--based risk prediction models are deployed in real-world settings, continuous monitoring is essential for detecting model degradation in a timely manner. However, in many settings, model predictions (e.g., clinical risk scores) guide decision-making and, in turn, affect the future data available for prediction. This feedback loop, known as performativity, can bias standard measures of model performance, rendering naive model evaluation and monitoring procedures invalid. In this work, we develop novel statistical methods for monitoring deployed prediction models under performativity. Adopting a causal inference framework, we distinguish between the observed data distribution and the counterfactual distribution, and introduce a monitoring criterion based on (stratified) counterfactual model calibration. We then construct new test statistics for monitoring counterfactual model calibration based on the theory of test martingales and develop nonparametric model monitoring procedures under performativity. We establish asymptotic properties for the proposed monitoring statistics under an idealized setting in which the calibration curve can be consistently estimated, and provide two monitoring procedures for practical applications where consistency may be infeasible. Through extensive simulation studies and semi-synthetic clinical data analysis, we demonstrate that the proposed procedures exhibit well-controlled false-alarm behavior while maintaining robust performance in detecting model degradation across a wide range of settings.

非线性动力学分会场

Hebbian与反Hebbian学习规则驱动的复杂网络自适应同步演化

袁五届　淮北师范大学

摘要：受神经系统Hebbian学习规则（一种广泛的突触可塑性）的启发，我们研究了网络的同步动力学反馈作用于网络连接权的一种自适应策略，这种策略可以使网络最终演化出大多真实网络所具有的社团结构并呈现出稳定的社团动力学同步簇，特别地，这种自适应策略可以有效地用来检测网络的弱社团结构。另外，受神经系统反Hebbian学习规则（一种“反常”的突触可塑性）的启发，我们研究了网络的动力学同步通过连接新边来反馈作用于网络的连接结构，大量的数值模拟表明，这种自适应网络最终会演化到稳定的同步态，并自发地形成很多真实网络所具有的无标度结构。

双周期驱动重塑生物振荡系统的阿诺德舌与同步转变

徐飞　安徽师范大学

摘要：生物信号网络是典型的远离平衡态复杂系统，其内部的时间尺度分离和反馈调控能够产生丰富的振荡、同步及状态转变等动力学行为。外部时间依赖驱动如何与系统内禀时间尺度相互作用，并进一步重塑非平衡系统中的有序行为和同步转变，是非线性动力学与复杂系统研究中的重要问题。本报告以p38 MAPK免疫信号转导网络为快慢生物振荡器模型，考察IL-1β和Dox双周期驱动下的同步动力学。通过旋转数分析，揭示双周期驱动对系统同步结构的调控作用。结果表明，双周期驱动可拓展1:1相位锁定区域并抑制同步间隙，而驱动幅值与频率的非对称性可诱导相位锁定、拟周期振荡与混沌之间的转变。结合相位响应曲线与相位约化理论，进一步阐明频率失谐与有效耦合强度对同步边界及共振结构的影响。本研究为理解多时间尺度非平衡系统中的同步与复杂动力学转变提供了新的理论视角。

布朗运动局部时间大偏差理论和维度诱导的动力学相变

陈含爽　安徽大学

摘要：随机系统的涨落研究是非平衡统计力学与概率论的核心内容，其应用涵盖非平衡相变、随机热力学、群体动力学等领域；其中 “大偏差（稀有事件）” 是近几十年的热点研究方向，随机变量的涨落性质可完全编码于大偏差函数中。大偏差框架下最显著的现象之一是动力学相变，并由大偏差函数的奇异性（非解析性）定义，已在格气模型、驱动扩散系统、玻璃动力学约束模型等多自由度系统中被观测到；在少自由度系统中，动力学相变通常出现于随机动力学的弱噪声极限，近年也在随机谐振子、布朗运动等简单模型中被报道。本报告聚焦d维布朗粒子在单位半径球壳处的局域时间密度（定义为，为粒子到原点的径向距离）涨落性质。在观测时间的极限下，局域时间密度满足大偏差原理：当时，速率函数全空间解析；当时，会在特定点（仅依赖空间维度）处出现非解析性。这种奇异性标志着四维以上空间中发生一级动力学相变，该相变伴随布朗轨迹大偏差过程中的时间相位分离现象。最后，通过稀有事件模拟方法验证了上述理论结果。

无序耦合系统的通用动力学平均场形式-基于路径积分方法

刘聪　安庆师范大学

摘要：将具有无序耦合的多粒子集体动力学，约化为波动环境中单粒子行为的动力学平均场方法，为我们理解高维复杂耦合系统提供了关键理论支撑。然而是否存在不依赖具体耦合形式的通用平均场结果尚不清楚。基于路径积分方法，我们推导出动力学平均场方程的通用表达式及其高阶关联。数值模拟结果进一步显示通用的动力学平均场表达式适用于扩散耦合、乘性耦合、以及相耦合等多种形式中。并且该表达式也可被用于刻画由淬火无序引发的混沌转变、混沌边缘共振等现象。

计算流体力学分会场

Learning Explicit Symbolic Smoothness Indicators for WENO Schemes via Kolmogorov–Arnold Networks

程自强　合肥工业大学

摘要：This work introduces an interpretable, KAN-based framework to learn data-driven β-type smoothness indicators for WENO schemes. Unlike typical black-box MLPs, the structural interpretability of KANs allows the trained network to be converted into explicit closed-form analytical formulas. Consequently, the new indicators can be coded directly into standard solvers, eliminating runtime neural-network inference overhead. While preserving the classical WENO weighting architecture, numerical simulations for hyperbolic conservation laws show that the proposed scheme achieves sharper shock resolution and superior accuracy compared to WENO-Z and WENO-NN. Furthermore, it significantly outperforms WENO-NN in computational speed, offering a robust and efficient approach to enhancing high-order shock-capturing methods via symbolic learning.

Convergence of the Hybrid WENO Scheme for Steady Compressible Navier–Stokes Equations in Curved Geometries Using Cartesian Grids

宛一飞　安徽大学

摘要：Classical weighted essentially non-oscillatory (WENO) and other nonlinear schemes often face challenges in achieving steady-state convergence, although substantial progress has already been made in the simulations of compressible Euler equations, there are few contributions for the steady-state simulations of compressible Navier–Stokes (NS) equations. To address this issue, we adopt the fifth-order hybrid WENO (WENO-H) finite difference scheme, designed to achieve machine-zero residual in numerical iterations. The WENO-H scheme employs fifth-order linear reconstruction in smooth regions, guided by an effective smoothness detector, and smoothly degrades to lower-order reconstruction to prevent non-physical oscillations. It ensures seamless transitions from smooth to discontinuous regions through a smoothing transition zone. In solving compressible NS equations in curved geometries on Cartesian grids, ghost point values outside the physical domain are determined using the fifth-order WENO extrapolation method coupled with simplified inverse Lax-Wendroff procedures. Two sets of ghost point values are utilized to handle convective and diffusive terms discretization near boundaries. Furthermore, the pressure term in primitive variables is substituted with the temperature term to facilitate the imposition of the adiabatic boundary condition. A transitional interpolation technique is proposed to enhance steady-state convergence near free-stream boundaries. Numerical experiments demonstrate that the WENO-H scheme achieves robust steady-state convergence across extensive benchmark examples of NS equations in curved geometries. The scheme exhibits good non-oscillatory property, particularly in scenarios involving strong discontinuities. Moreover, it provides high resolution for both steady and unsteady problems containing multi-scale structures.

Modeling and Analysis of Mixed Convection and Heat Transfer of Space-Fractional Viscoelastic Fluids

赵金虎　阜阳师范大学

摘要：Finite volume method is developed for the mixed convection boundary layer flow and heat transfer of viscoelastic fluid over a flat plate. The spatial fractional derivative of the Riemann-Liouville type is employed in the constitutive relation and modified Fourier’s law respectively. Nonlinear and coupled boundary layer governing equations are formulated with non-uniform boundary conditions. The discretized scheme combined with the shifted Grünwald-Letnikov formula is proved to be conditionally stable, further the convergence and accuracy of the numerical solutions are presented. Results demonstrate that the space fractional derivative parameters have strong effects on the velocity and temperature distributions.

LDG methods for the time-dependent Ginzburg–Landau equations

王昊　安徽师范大学

摘要：We propose a class of high-order, decoupled, linearized LDG methods for the time-dependent Ginzburg–Landau equations. The LDG formulation can handle discontinuous polynomial approximations and nonsmooth domains without a Hodge decomposition. We prove an exact energy-dissipation law in the temporal gauge and identify a projection- compatibility obstruction in the Lorentz gauge caused by the φψ coupling. Numerical experiments show that the observed accuracy depends on polynomial degree, mesh geometry, and flux choice; discrete gauge invariance on nonsmooth domains is problem dependent and may degrade significantly on a multiply connected nonconvex domain. Vortex simulations confirm the method's ability to resolve complex superconducting states.

交通分会场

大数据与AI驱动的交通精准感知、网络优化与仿真实景化研究

肖贇　合肥大学

摘要：针对交通系统强非线性特征，报告融合卡口、手机信令、视频等多源数据，结合非线性聚类与深度学习方法，实现交通流、客流、人流的精准识别与时态监测，拟解决传统观测数据稀疏、多主体混叠难题。基于精细化感知结果，从物流网络、公交线网、动态运力、关键节点等方面开展运输网络自适应优化。针对SUMO传统仿真偏向动画、与真实场景差异大的痛点，利用AI视频提取真实人车行为对仿真模型进行校准，提升仿真真实性。该方法可为城市交通管控、规划与数字孪生建设提供理论与技术支撑。

城市低空物流-公交车载光伏补能系统的管理与优化

胡茂彬　中国科学技术大学

摘要：城市低空物流的规模化运行不仅受机载电池容量限制，也受到补能资源时空可达性的约束。现有公共交通辅助无人机配送的研究通常将公交车视为移动交通资源或理想化移动充电节点，而较少刻画公交车的车顶光伏出力、辅助储能能力及其在天气变化下的可用补能能力。我们提出一种面向公交协同城市无人机物流的气象驱动公交移动光伏补能框架，其中装配车顶光伏和独立小型辅助储能的公交车按照既定线路与时刻表运行，并作为移动能源节点为无人机提供有限机会式补能。基于交通大数据和强化学习方法，构建公交线路、站点、无人机状态和移动补能节点耦合的扩展交通-能源模型，描述无人机飞行、公交协同转移、站点等待、机会式充电、回仓补能和电量状态更新过程。

Traffic Dynamics of Vehicles Passing Each Other on Bidirectional Undivided Narrow Roads

郭宁　合肥工业大学

摘要：This study investigates traffic dynamics on bidirectional undivided narrow roads. A novel heuristic-based passing (HP) model is proposed to microscopically model the complex passing maneuvers. Considering both perceived visual stimuli and steering imprecision the model incorporates two heuristics that determine the driving direction and speed in the navigation of the most direct but unobstructed route. Real-world experiments analyzing the passing process of two vehicles on undivided narrow roads with varying widths were conducted. The results reveal that vehicles decelerated, veered toward the roadside to facilitate an oncoming vehicle’s passage and then readjusted to the center of the road, providing empirical support to the proposed HP model. In addition, as the road width increased, both the passing speed and clearance distance increased. The calibrated HP model could replicate the trajectories of the passing vehicles, demonstrating its accuracy. Real-world field observations were made at two undivided roads of varying widths and opposing traffic densities to investigate the macroscopic traffic patterns under the influence of lateral friction. The results showed that, on the undivided narrow road, when the traffic density in the travel direction was relatively low, the presence of high-density opposing traffic significantly reduced the speed of vehicles in the travel direction. Moreover, the proposed microscopic HP model well replicated the macroscopic traffic flow patterns. Overall, the results of this study enhance the understanding of traffic dynamics in the passing process on undivided narrow roads, offer insights for local road geometric design, and help identify sources of congestion evolution.

Dynamic Joint Routing and Platooning for AutonomousTrucks under Uncertainty

陈书恺　合肥工业大学

摘要：This study investigates a joint routing and platooning problem for autonomous trucks (ATs) under uncertainty. As the number of ATs that receive transportation requests is uncertain, operators should periodically decide AT routing and platoon formation during the operation horizon so that the maximum number of platoons can be formed without causing excessive detour and waiting cost. We formulate the problem as an infinite-horizon Markov decision process (MDP) that aims to minimize expected operation cost subject to the constraints of delivery deadline and the maximum platoon length. To address the curses of dimensionality of MDP, we propose two state aggregation schemes and a tailored approximate dynamic programming (ADP) algorithm. The developed algorithm adopts a customized dual-head neural network (DHNN) learning framework that approximates nonlinear value function and accelerates one-period problem solving through linear value-function surrogate. Extensive numerical experiments on a variety of networks demonstrate the superior performance of proposed methods against heuristic policies and various benchmarks of ADP with linear models and NN.

量子多体分会场

凝聚态物理的标准模型：从朗道费米理论到拓扑场论

胡森　中国科学技术大学

摘要：凝聚态物理有丰富复杂的现象，有无粒子物理一样的标准模型。对于二维材料，我们提出拓扑场论是其理论基础。

拓扑弦理论的进展

黄民信　中国科学技术大学

摘要：我们简要介绍拓扑弦理论的来源和研究现状，然后讨论一些和报告者相关的进展，包括高亏格配分函数的计算，和代数几何理论的联系，和一类量子力学系统的精确能谱的联系。

从朗道范式到范畴对称性：量子多体相与相变的新语言

张智浩　中国科学技术大学

摘要：理解相与相变是量子多体的核心问题。朗道对称破缺理论获得了极大的成功，但在拓扑序等新奇的物理系统中仍有一定的局限性。本报告将介绍一种新视角：当局部可观测量消失，不同物理系统之间的“关系”成为关键。这些关系包含边界、畴壁、拓扑缺陷，可用范畴语言刻画，并被视为广义对称性。体边关系与拓扑全息进一步表明，对称性等价于高一维的拓扑序。这些思想虽然是在研究拓扑序中产生的，但也可应用于更一般的无能隙多体系统的研究。

中子星内部的量子多体效应：超流量子临界性

朱豪富　中国科学技术大学

摘要：中子星内部的核物质是强关联量子多体系统，量子多体效应对中子星的物态方程、热演化等物理量产生重要影响。中子星的传统热演化理论在解释Cassiopeia A中子星快速冷却现象时面临困难。我们近期提出，中子星内部可能存在超流量子临界现象，计算发现该量子多体效应使核子物质呈现非费米液体行为，导致核子热容量与中微子发射率出现对数型温度依赖关系。将这些量子多体效应纳入热演化模拟，能够解释Cassiopeia A中子星的快速冷却天文观测结果。

AI for Science分会场

Community Detection in Brain Connectivity Networks via Functional Data

陈书延　中国科学技术大学

摘要：Community recovery from discretely observed noisy functional data is challenging when weak dependence is obscured by network estimation error. We develop a joint estimation framework for community recovery from resting-state functional magnetic resonance imaging data that simultaneously estimates functional connectivity networks and latent community structure. The method models regional signals as noisy functional observations and integrates community assignment directly into network estimation, thereby leveraging weak coordinated dependence through shared structural information and mitigating error propagation from intermediate estimation. We establish estimation error bounds for the precision matrices and consistency of community recovery. Simulation studies demonstrate improved recovery under weak dependence and noisy discrete observation. An application to Alzheimer’s disease neuroimaging data reveals structured differences in temporal and limbic connectivity that are less apparent under existing approaches.

Learning Feynman Integrals

刘元彻　中国科学技术大学

摘要：Feynman-integral calculations involve two complementary challenges: reconstructing exact analytic structures from numerical data and finding a basis in which those structures become manifest. I will present two AI-assisted approaches. First, symbolic regression is used to infer compact exact expressions from numerical canonical differential equations, with every proposal verified algebraically. Second, I will introduce CANON, a stateful AI-agent framework that coordinates IBP reduction, differential equations, and several strategies for generating uniform-transcendental candidates. The search proceeds sector by sector, while accepted results, failed candidates, and diagnostics are recorded in a persistent scientific state.

单细胞数据的统计分析——癌症治疗靶点的发掘与验证

赵方杰   中科院吴文俊数学重点实验室

摘要：肿瘤的内在异质性与复杂的肿瘤微环境是导致治疗抵抗和预后差异的重要因素，而传统批量测序方法难以解析细胞类型间的表达差异，制约了功能性治疗靶点的精准发现。单细胞RNA测序结合公共数据库资源，为在单细胞分辨率下解构肿瘤生态系统、识别关键致病细胞亚群提供了可推广的研究路径。本报告以小细胞肺癌的公开单细胞数据为实例，概述一套“数据整合—统计分析—靶点验证”方法框架。该方法可为多种癌症的靶点筛选与精准治疗研究提供参考，并可借助大语言模型辅助分析流程自动化，提高从公共单细胞大数据中筛选候选治疗靶点的效率。