第四届网络经济博弈论坛

数字经济时代，大语言模型为教育领域的创新提供了新动力。本报告以多学科融合教育为核心，系统讨论大语言模型的垂直开发策略。首先，从技术原理出发，分析其强大的知识吸纳能力和沟通表达能力。其次，概述大语言模型在产业应用中的生态体系，从资源整合到实践案例，阐明其经济价值。接着，介绍垂直开发的技术架构，展示数据定制、微调和高效部署等具体环节。进一步，探讨人机分工的协作模式，分析模型与领域专家在教育场景中的互补性。报告还将梳理可信性与可解释性在产品开发中的必要性，提供面向教育领域的优化策略。最后，通过实践效果与未来展望，尝试探讨大模型垂直开发如何驱动数字经济教育的长远发展。

赵琳，上海财经大学教授，在清华大学数学系取得理学学士学位、理学博士学位。主要研究方向为行为管理、风险管理、运营管理，研究成果发表于管理学、应用经济学等国内外期刊，2019年至今任Management Science决策分析领域Associate Editor。

We study a set function via its market extension: each player in the set possesses certain endowments and the worth of a subset is determined by the corresponding total endowments, as well as a common (continuous) production function. Compared with usual extensions where each production function corresponds to a single set function, each production in the market extension corresponds to an infinite number of set functions. We prove that the associated set functions are supermodular if and only if the production function is ultramodular. We study Walrasian cores of ultramodular production functions and prove several desirable properties resembling well-known ones of supermodular set functions. (joint with Dong Liang and Ning Yu)

曹志刚，北京交通大学经济管理学院教授。长期从事合作博弈、交通博弈、 网络博弈和算法博弈等方面的研究，在包括Operations Research、Mathematics of Operations Research、 Games and Economic Behavior和《中国科学：数学》在内的期刊上发表多篇论文。 相关成果曾获中国信息经济学理论贡献奖、系统科学与系统工程青年科技奖、 中国决策科学青年科技奖和关肇直青年研究奖等荣誉。先后主持国家自然科学基金委的青年、面上和优青项目。

This paper analyzes a network game in a local-average setup where a player's payoff depends on the social norm he confronts. I focus on an optimal targeting intervention problem where a planner seeks to maximize his objective via changing the individual attributes and/or restructure the underlying network subject to certain constraints. First, I discuss the absolute characteristic intervention problem under a quadratic budget constraint. I give the limit forms of the optimal intervention under, respectively, extremely large and extremely small budgets. Then, I proceed to identify the errors stemming from approximating the optimal intervention with its limit forms. Next, I study the relative intervention problem in which a planner maximizes the ratio of the equilibrium social welfare over its first-best counterpart subject to a quadratic budget constraint. In what follows, I turn to the structural intervention problem, where a social planner designs endogenously a network to maximize the social welfare subject to the unanimous consent constraints of players. It is shown that whether or not a positive result is achievable relies on the degree of individual conformism, and their tolerance regarding minor loss.

Dawen Meng, a professor at the School of Economics, Shanghai University of Finance and Economics, specializes in game theory, mechanism design theory, and social network theory. He has published over ten papers in prestigious journals, including Games and Economic Behavior(*2), Economic Theory, Review of Economic Design, Operations Research Letters, China Economic Quarterly(*2), and China Industrial Economics, as well as in renowned conferences such as WINE. He has also led two projects funded by the National Natural Science Foundation of China.

Motivated by practical constraints in online advertising, we investigate single-parameter auction design for bidders with constraints on their Return On Investment (ROI) – a targeted minimum ratio between the obtained value and the payment. We focus on ex post ROI constraints, which require the ROI condition to be satisfied for every realized value profile. With ROI-constrained bidders, we provide a full characterization of the allocation and payment rules of dominant-strategy incentive compatible (DSIC) auctions. In particular, we show that given any monotone allocation rule, the corresponding DSIC payment should be the Myerson payment with a rebate for each bidder to meet their ROI constraints. Furthermore, we also determine the optimal auction structure when the item is sold to a single bidder under a mild regularity condition. This structure entails a randomized allocation scheme and a first-price payment rule, which differs from the deterministic Myerson auction and previous works on ex ante ROI constraints.

Xiaohui Bei is an Associate Professor in the Division of Mathematical Sciences at Nanyang Technological University. He obtained his Ph.D. from Tsinghua University in 2012. His research interests include topics in resource allocation, computational economics, and general algorithm design.

Understanding human strategic behavior is crucial for designing effective market mechanisms. Conventional economic models often rely on assumptions of human rationality and perfect knowledge, which sometimes make it difficult to accurately characterize human behavior in complex real-world markets. In this talk, we will introduce two deep learning-based frameworks for learning human behavior. First, we study scenarios where closed-form economic models are available, and propose a behavior learning framework that integrates the economic models with deep learning. Second, we study scenarios where closed-form economic models are unavailable, and propose a framework that incorporates economic properties and constraints into behavior learning. Using penny auctions and continuous double auctions as case studies, we will demonstrate the experimental performance of our frameworks on real human behavior data.

Haoran Yu is currently an Associate Professor with the School of Computer Science and Technology, Beijing Institute of Technology. He received the Ph.D. degree from the Department of Information Engineering, the Chinese University of Hong Kong in 2016. From 2015 to 2016, he was a Visiting Student with the Yale Institute for Network Science and the Department of Electrical Engineering, Yale University. From 2018 to 2019, he was a Post-Doctoral Fellow with the Department of Electrical and Computer Engineering, Northwestern University. His current research interests lie in the interdisciplinary area between game theory and artificial intelligence, with a focus on real human strategic behavior analysis. His work has been published mainly in conferences and journals within the fields of artificial intelligence and networking (e.g., AAAI, IJCAI, ACM SIGMETRICS, ACM MobiHoc, IEEE INFOCOM, IEEE/ACM TON, IEEE JSAC, and IEEE TMC).

近年来，安全博弈模型受到了大量关注，也被应用于不少实际场景中，如保护野生动物、防止城市犯罪等。但很少有研究考虑举报人在安全博弈中所起的作用。我们将举报人这一角色引入标准安全博弈模型，并系统性地研究了策略性举报人对安全博弈的影响。

我们将该问题建模为一个三方博弈，并认为举报人具有一个未知的类型，该类型将决定举报人的策略行为。在只有一个举报人的场景，我们定义了一个新的均衡概念，并借助机制设计相关理论，刻画了最优解的形式，证明最优策略中所使用的信息数量与目标数量相同，并由此简化了最优策略的计算问题。在具有多个举报人的场景，我们设计了一个基于简单多数投票的信息汇总策略，并证明该策略是最优的，且具有高效的算法。

沈蔚然是中国人民大学高瓴人工智能学院准聘副教授，本科毕业于清华大学电子工程系，2019年于清华大学交叉信息研究院获博士学位，2019年至2020年于卡内基梅隆大学担任博士后研究员。主要研究方向为多智能体系统、博弈论、机制设计和机器学习，在相关领域国际会议发表高水平论文三十余篇，并担任多个国际会议的高级程序委员会委员及领域主席。在机制设计方面的研究成果已在百度、字节跳动等互联网平台落地实现。

Many areas such as algorithmic game theory, online algorithms, and operation research switched their focus from the worst-case analysis to stochastic optimization frameworks. This shift is well justified with the rapid growth and easy accessibility of data and the desire of many online companies to leverage statistical analysis. However, despite the unprecedented amounts of data, the dimensionality of data has also significantly increased. E.g., a prototypical example of Bayesian mechanism design for single-item auction with $n$ buyers may have $n$ in the range of hundreds in the online advertising context. In the stochastic optimization, it is typically assumed that the prior distributions are explicitly given by an exact probabilistic model, for the most part an independent multi-dimensional distribution. The later is a strong assumption that is extremely hard to verify statistically. In this talk, we discuss a new approach based on robust optimization. The idea is to relax strong mutual independence to a more statistically friendly pair-wise or more generally $k$-wise independence assumption. We then discuss how a few commonly used mechanisms such as sequential posted pricing and celebrated Myerson's mechanism, that maximizes revenue for the mutually independent prior, perform in this new framework.

Based on joint work with Zhiqi Wang [FOCS 24] and with Ioannis Caragiannis, Pinyan Lu, and Zihe Wang [WINE 21].

Nick Gravin is a professor of computer science at Shanghai University of Finance and Economics in the key national laboratory of Interdisciplinary Research on Computation and Economics. His primary research interests are in algorithmic mechanism design, stochastic optimization, and learning theory. Dr. Gravin has received Ying Xian Scholar title, WINE best paper award, and Microsoft fellowship.

We initiate the study of confusion matrix design. In this problem, an algorithm designer needs to generate a machine learning model (for a classification task from contexts to labels) which makes predictions for a population of downstream decision makers. The prediction accuracy of the machine learning model is characterized by its confusion matrix, which is a stochastic matrix where each entry encodes the probability of predicting the true label to another label. Each downstream decision maker faces a separate optimization task and will decide her binary action based on her own context, realized prediction given her context, and the confusion matrix selected by the algorithm designer. Decision makers are heterogeneous, as they may hold different contexts. Both the decision makers and the algorithm designer will obtain utilities that are determined by the actions the decision makers take, and their true labels. The goal of the algorithm designer is to design a single confusion matrix that is used for all decision makers subject to some feasibility constraints in order to maximize his net utility. We consider a general class of net utility functions, which could be a combination of both decision makers' utilities and the algorithm designer's utilities. Classic outcome-independent utility and utilitarian/Nash/egalitarian social welfare are all special cases of our net utility formulation.

We study the above problem through an information design framework, where we view training machine learning model as designing an information structure (signaling scheme) subject to some specific constraints motivated by the machine learning literature. By building the connection to the public persuasion with heterogeneous priors, we design convex programming-based algorithms that compute the optimal confusion matrix subject to (i) post-processing constraints and (ii) receiver operating characteristic (ROC) constraints in polynomial time, respectively. Besides the computational results, we also obtain analytical structural results and numerical results for the special cases of outcome-independent utility and social-aware utility, by utilizing the convex programming-based characterization of the optimal confusion matrix.

This talk is based on the joint work with Wei Tang (CUHK DOT). The preliminary conference version will appear in ITCS 2025.

Yiding Feng is an assistant professor at HKUST IEDA. Previously, he worked as a principal researcher at the University of Chicago Booth School of Business, and a postdoctoral researcher at Microsoft Research New England. He received his Ph.D. from the Department of Computer Science at Northwestern University in 2021, and his BS degree from ACM Honors Class at Shanghai Jiao Tong University in 2016. His research focuses on operations research, economics & computation, and theoretical computer science. He was the recipient of the INFORMS Auctions and Market Design (AMD) Michael H. Rothkopf Junior Researcher Paper Prize (second place), and the APORS Young Researcher Best Paper Award.

Clock auctions are a natural class of simple auction mechanisms. Clock auctions are known to enjoy strong properties, such as obvious strategyproofness, credibility, and privacy, which offer remarkable robustness in real-world scenarios. In contrast, computational aspects of clock auctions have not been explored as deeply. In this work, we investigate the computational problems of (1) checking whether a given way of allocating items to buyers (i.e., an allocation function) can be implemented by clock auctions, (2) finding a clock auction protocol (whenever there exists one) that implements a given allocation function, and (3) optimizing social welfare using clock auctions. We give polynomial-time algorithms for tasks (1) and (2), and show that task (3) is NP-hard. En route, we derive a complete characterization of the class of allocation functions that can be implemented by clock auctions.

On the other hand, we also investigate the economic efficiency of clock auctions. We connect welfare maximization (particularly in the special case with "independent groups") to the much cleaner problem of "upper tail extraction". We show that the (in)existence of constant-factor clock auction protocols for either problem implies that for the other problem. This can be viewed as a framework for designing approximately efficient clock auctions, or proving impossibility results thereof. To illustrate the power of this framework, we establish the existence of constant-factor clock auctions for upper tail extraction with iid agents, which implies the existence of constant-factor clock auctions for welfare maximization with independent groups that are each "homogeneous". We view this as strong evidence that constant-factor clock auctions exist for the general problem.

Hanrui is an Assistant Professor in Computer Science and Engineering at the Chinese University of Hong Kong. He got his BE from Tsinghua University, and his PhD from Carnegie Mellon University. His research focuses on Economics and Computation – problems with economic motivations that can be approached using techniques from computer science.

In cooperative games, we study how values created or costs incurred by a coalition are shared among the members within it, and the players may join the coalition in a online manner such as investors invest a startup. Recently, Ge et al. [10] proposed a new property called incentives for early arrival (I4EA) in such games, which says that the online allocation of values or costs should incentivize agents to join early in order to prevent mutual strategic waiting. Ideally, the allocation should also be fair, so that agents arriving in an order uniformly at random should expect to get/pay their Shapley values. Ge et al. [10] showed that not all monotone value functions admit such mechanisms in online value sharing games. In this work, we show a sharp contrast in online cost sharing games. We construct a mechanism with all the properties mentioned above, for every monotone cost function. To achieve this, we first solve 0-1 valued cost sharing games with a novel mechanism called Shapley-fair shuffle cost sharing mechanism (SFS-CS), and then extend SFS-CS to a family called generalized Shapley-fair shuffle cost sharing mechanisms (GSFS-CS). The critical technique we invented here is a mapping from one arrival order to another order so that we can directly apply marginal cost allocation on the shuffled orders to satisfy the properties. Finally, we solve general valued cost functions, by decomposing them into 0-1 valued functions in an online fashion.

赵登吉，上海科技大学信息学院常任副教授/博导，信息学院机器人中心主任。国际多智能体系统基金会（IFAAMAS）董事会董事，多智能体旗舰期刊JAAMAS副主编。CCFAI专委多智能体系统学组秘书长，CCF YOCSEF上海副主席。上海市优秀教学成果一等奖、一流课程、和重点课程获得者。他开创并推动了基于社交网络的机制设计和在线合作博弈中的激励早加入两项重要的创新研究，相关成果获得了AAMAS 2024等四项国际顶会最佳论文奖。他也受邀在 IJCAI 2022上发表Early Career Spotlight演讲。

In the field of online sequential decision-making, we address the problem with delays utilizing the framework of online convex optimization (OCO), where the feedback of a decision can arrive with an unknown delay. Unlike previous research that is limited to Euclidean norm and gradient information, we propose three families of delayed algorithms based on approximate solutions to handle different types of received feedback. Our proposed algorithms are versatile and applicable to universal norms. For each type of algorithm, we provide corresponding regret bounds under cases of general convexity and relative strong convexity, respectively. We also demonstrate the efficiency of each algorithm under different norms through concrete examples. Furthermore, our theoretical results are consistent with the current best bounds when degenerated to standard settings.

刘正阳，北京理工大学计算机学院副研究员，硕士生导师。研究方向算法博弈论和在线学习，在顶级会议STOC、CCC、WWW、AAAI和AMAAS上发表论文10余篇，主持国家自然科学基金青年项目和面上项目各1项。目前为全国理论计算机科学委员会和计算经济学专业组执行委员。

The predominant setting in classic auction theory considers bidders as \emph{utility maximizers (UMs)}, who aim to maximize quasi-linear utility functions. Recent autobidding strategies in online advertising have sparked interest in auction design with \emph{value maximizers (VMs)}, who aim to maximize the total value obtained. In this work, we investigate revenue-maximizing auction design for selling a single item to a mix of UMs and VMs. Crucially, we assume the UM/VM type is private information of a bidder. This shift to a multi-parameter domain complicates the design of incentive compatible mechanisms. Under this setting, we first characterize the optimal auction structure for auctions with a single bidder. We observe that the optimal auction moves gradually from a first-price auction to a Myerson auction as the probability of the bidder being a UM increases from 0 to 1. We also extend our study to multi-bidder setting and present an algorithm for deriving the optimal lookahead auction with multiple mixed types of bidders.

Xiang Yan is a researcher in Huawei Taylor Lab. He received the PhD degree from Department of Computer Science of Shanghai Jiao Tong University, advised by Prof Xiaotie Deng. He was also visiting scholars Harvard University and Hong Kong University of Science and Technology. His research interest has been mainly in algorithmic game theory. In particular, in following topics: game theory and mechanism design, Internet and computational economics, crowd sourcing, reinforcement learning and multi-agent systems.

This paper studies a joint design problem where a seller can design both the signal structures for the agents to learn their values, and the allocation and payment rules for selling the item. In his seminal work, Myerson (1981) shows how to design the optimal auction with exogenous signals. We show that the problem becomes NP-hard when the seller also has the ability to design the signal structures. Our main result is a polynomial-time approximation scheme (PTAS) for computing the optimal joint design with at most an $\epsilon$ multiplicative loss in expected revenue. Moreover, we show that in our joint design problem, the seller can significantly reduce the information rent of the agents by providing partial information, which ensures a revenue that is at least $1−1/e$ of the optimal welfare for all valuation distributions.

Yingkai Li is an Assistant Professor (Presidential Young Professor) in Economics at the National University of Singapore. He was a postdoctoral fellow at the Cowles Foundation for Research in Economics and the Department of Computer Science at Yale University, working with Prof. Dirk Bergemann and Prof. Yang Cai. He received his PhD in Computer Science from Northwestern University, advised by Prof. Jason Hartline. He completed his BS in Computer Science from Shanghai Jiaotong University in 2015 and his MS in Computer Science from Stony Brook University in 2018.