arXiv subjects

Ningyuan Li

Publications and source records attributed to Ningyuan Li.

16 recordsLinked to original sources

arXiv2026

TrustMargin: Training-Free Arbitration between Parametric Memory and Retrieved Evidence in Large Language Models

Large language models answer knowledge-intensive questions using both parametric memory and retrieved evidence, but neither source is uniformly reliable. Retrieval can fill knowledge gaps, yet distracting passages may override correct closed-book answers. We study this post-generation conflict as answer-level source arbitration: given Direct and RAG answers from the same frozen model, decide which source to trust. We propose TRUSTMARGIN, a training-free, plug-and-play arbitration layer that scores the two existing candidates with the model's own likelihoods. It combines a parametric-prior margin, which tests whether memory accepts the retrieved answer, with an evidence-binding margin, which discounts passage-only salience and measures question-specific support. TRUSTMARGIN selects between Direct and RAG without fine-tuning, external judges, or additional generation. Across 2WIKIMQA and CWQA with three LLaMA scales, TRUSTMARGIN consistently improves over Direct generation and BM25-RAG, recovers part of the Direct/RAG oracle gap, and generalizes to multiple training-free RAG pipelines.

cs.CL↗

arXiv2026

Understanding the Ability of LLMs to Handle Character-Level Perturbation

This work investigates the resilience of contemporary large language models (LLMs) against frequent character-level perturbations. We examine three types of character-level perturbations including introducing numerous typos within words, shuffling the characters in each word, and inserting a large number of invisible characters into the text. Surprisingly, even under severe perturbation, such as shuffling nearly all words character-wise to produce text that is almost unreadable to humans, or inserting invisible characters which are several times more than the visible ones as noise, many LLMs still maintain notable performance. We explore the underlying causes of this robustness and find that LLMs exhibit remarkable resilience to chaotic segmentation and fragmented tokenization. Furthermore, we examine the mechanisms by which LLMs remove perturbations to correctly comprehend text, including both implicit and explicit mechanisms for character-level perturbation. We hope that our findings on the low-level robustness of LLMs will unveil their inherent architectural strengths, reveal the potential risks of their misuse, and inform the reliable deployment of LLMs across diverse application scenarios.

cs.CL↗

arXiv2026

SGR-Bench: Benchmarking Search Agents on State-Gated Retrieval

Recent advances in large language models and tool-using agents have expanded the range of benchmarked web tasks. Yet an important class of specialized retrieval tasks remains undercharacterized. On many specialized data-retrieval websites, answer-bearing evidence becomes accessible only after establishing the correct site-specific retrieval state through filters, views, hierarchies, or scopes. We term this capability state-gated retrieval (SGR). We introduce SGR-Bench, a benchmark for this setting containing 100 expert-curated tasks spanning six source families and 12 public data ecosystems. Each task requires discovering the appropriate website and configuring its site-specific retrieval state to produce a structured answer. SGR-Bench pairs constraint-guided and goal-oriented formulations of the same underlying problems, enabling controlled comparisons between explicit and implicit guidance for state-gated retrieval. We evaluate eight CLI-based agentic LLM systems and three commercial search-agent products. On SGR-Bench, the strongest system reaches only 66.18% item-level F1, while row-level F1 remains much lower. A manual audit of 156 analyzable failed CLI trajectories shows why: agents often reach a relevant web source, but establish the wrong site-specific retrieval state. Retrieval-scope drift (37.2%) and criterion mismatch (27.6%) dominate, whereas final answer composition accounts for only 10.3%. The dataset and single-case evaluation instructions are available at https://huggingface.co/datasets/PKUAIWeb/SGR-BENCH.

cs.AI↗

arXiv2026

Incentive Effects of a Cut-Off Score: Optimal Contest Design with Transparent Pre-Selection

Shortlisting is a common and effective method for pre-selecting participants in competitive settings. To ensure fairness, a cut-off score is typically announced, allowing only contestants who exceed it to enter the contest, while others are eliminated. In this paper, we study rank-order contests with shortlisting and cut-off score disclosure. We fully characterize the equilibrium behavior of shortlisted contestants for any given prize structure and shortlist size. We examine two objective functions: the highest individual performance and total performance. For both objectives, the optimal contest is in a winner-take-all format. For the highest individual performance, the optimal shortlist size is exactly two contestants, but, in contrast, for total performance, the shortlist size does not affect the outcome, i.e., any size yields the same total performance. Furthermore, we compare the highest individual performance achieved with and without shortlisting, and show that the former is 4/3 times greater than the latter.

cs.GT↗

arXiv2026

PankRAG: Enhancing Graph Retrieval via Globally Aware Query Resolution and Dependency-Aware Reranking Mechanism

Recent graph-based RAG approaches leverage knowledge graphs by extracting entities from a query to fetch their associated relationships and metadata. However, relying solely on entity extraction often results in the misinterpretation or omission of latent critical information and relationships. This can lead to the retrieval of irrelevant or contradictory content, as well as the exclusion of essential information, thereby increasing hallucination risks and undermining the quality of generated responses. In this paper, we propose PankRAG, a framework designed to capture and resolve the latent relationships within complex queries that prior methods overlook. It achieves this through a synergistic combination of a globally-aware hierarchical resolution pathway and a dependency-aware reranking mechanism. PankRAG first generates a globally aware resolution pathway that captures parallel and progress relationships, guiding LLMs to resolve queries through a hierarchical reasoning path. Additionally, its dependency-aware reranking mechanism utilizes resolved sub-question dependencies to augment and validate the retrieved content of the current unresolved sub-question. Experimental results demonstrate that PankRAG consistently outperforms existing state-of-the-art methods, underscoring its generalizability.

cs.CL↗

arXiv2025

Ex-Ante Truthful Distribution-Reporting Mechanisms

This paper studies mechanism design for revenue maximization in a distribution-reporting setting, where the auctioneer does not know the buyers' true value distributions. Instead, each buyer reports and commits to a bid distribution in the ex-ante stage, which the auctioneer uses as input to the mechanism. Buyers strategically decide the reported distributions to maximize ex-ante utility, potentially deviating from their value distributions. As shown in previous work, classical prior-dependent mechanisms such as the Myerson auction fail to elicit truthful value distributions at the ex-ante stage, despite satisfying Bayesian incentive compatibility at the interim stage. We study the design of ex-ante incentive compatible mechanisms, and aim to maximize revenue in a prior-independent approximation framework. We introduce a family of threshold-augmented mechanisms, which ensures ex-ante incentive compatibility while boosting revenue through ex-ante thresholds. Based on these mechanisms, we construct the Peer-Max Mechanism, which achieves an either-or approximation guarantee for general non-identical distributions. Specifically, for any value distributions, its expected revenue either achieves a constant fraction of the optimal social welfare, or surpasses the second-price revenue by a constant fraction, where the constants depend on the number of buyers and a tunable parameter. We also provide an upper bound on the revenue achievable by any ex-ante incentive compatible mechanism, matching our lower bound up to a constant factor. Finally, we extend our approach to a setting where multiple units of identical items are sold to buyers with multi-unit demands.

cs.GT↗

arXiv2025

Beyond Advertising: Mechanism Design for Platform-Wide Marketing Service "QuanZhanTui"

On e-commerce platforms, sellers typically bid for impressions from ad traffic to promote their products. However, for most sellers, the majority of their sales come from organic traffic. Consequently, the relationship between their ad spending and total sales remains uncertain, resulting in operational inefficiency. To address this issue, e-commerce platforms have recently introduced a novel platform-wide marketing service known as QuanZhanTui, which has reportedly enhanced marketing efficiency for sellers and driven substantial revenue growth for platforms. QuanZhanTui allows sellers to bid for impressions from the platform's entire traffic to boost their total sales without compromising the platform's user experience. In this paper, we investigate the mechanism design problem that arises from QuanZhanTui. The problem is formulated as a multi-objective optimization to balance sellers' welfare and platform's user experience. We first introduce the stock-constrained value maximizer model, which reflects sellers' dual requirements on marketing efficiency and platform-wide ROI. Then, we propose the Liquid Payment Auction (LPA), an auction designed to optimize the balanced objectives while accounting for sellers' requirements in the auto-bidding environment. It employs a simple payment rule based on sellers' liquid welfare, providing a clearer link between their investment and total sales. Under mild assumptions, we theoretically prove desirable properties of LPA, such as optimality and incentive compatibility. Extensive experiments demonstrate LPA's superior performance over conventional auctions in QuanZhanTui.

cs.GT↗

arXiv2025

Optimal Prize Design in Parallel Rank-order Contests

This paper investigates a two-stage game-theoretical model with multiple parallel rank-order contests. In this model, each contest designer sets up a contest and determines the prize structure within a fixed budget in the first stage. Contestants choose which contest to participate in and exert costly effort to compete against other participants in the second stage. First, we fully characterize the symmetric Bayesian Nash equilibrium in the subgame of contestants, accounting for both contest selection and effort exertion, under any given prize structures. Notably, we find that, regardless of whether contestants know the number of participants in their chosen contest, the equilibrium remains unchanged in expectation. Next, we analyze the designers' strategies under two types of objective functions based on effort and participation, respectively. For a broad range of effort-based objectives, we demonstrate that the winner-takes-all prize structure-optimal in the single-contest setting-remains a dominant strategy for all designers. For the participation objective, which maximizes the number of participants surpassing a skill threshold, we show that the optimal prize structure is always a simple contest. Furthermore, the equilibrium among designers is computationally tractable when they share a common threshold.

cs.GT↗

arXiv2025

Pyramidal Flow Matching for Efficient Video Generative Modeling

Video generation requires modeling a vast spatiotemporal space, which demands significant computational resources and data usage. To reduce the complexity, the prevailing approaches employ a cascaded architecture to avoid direct training with full resolution latent. Despite reducing computational demands, the separate optimization of each sub-stage hinders knowledge sharing and sacrifices flexibility. This work introduces a unified pyramidal flow matching algorithm. It reinterprets the original denoising trajectory as a series of pyramid stages, where only the final stage operates at the full resolution, thereby enabling more efficient video generative modeling. Through our sophisticated design, the flows of different pyramid stages can be interlinked to maintain continuity. Moreover, we craft autoregressive video generation with a temporal pyramid to compress the full-resolution history. The entire framework can be optimized in an end-to-end manner and with a single unified Diffusion Transformer (DiT). Extensive experiments demonstrate that our method supports generating high-quality 5-second (up to 10-second) videos at 768p resolution and 24 FPS within 20.7k A100 GPU training hours. All code and models are open-sourced at https://pyramid-flow.github.io.

cs.CV↗

arXiv2025

Optimal Contest Design with Entry Restriction

This paper explores the design of contests involving $n$ contestants, focusing on how the designer decides on the number of contestants allowed and the prize structure with a fixed budget. We characterize the unique symmetric Bayesian Nash equilibrium of contestants and find the optimal contests design for the maximum individual effort objective and the total effort objective.

cs.GT↗

arXiv2024

Equilibrium Analysis of Customer Attraction Games

We introduce a game model called "customer attraction game" to demonstrate the competition among online content providers. In this model, customers exhibit interest in various topics. Each content provider selects one topic and benefits from the attracted customers. We investigate both symmetric and asymmetric settings involving agents and customers. In the symmetric setting, the existence of pure Nash equilibrium (PNE) is guaranteed, but finding a PNE is PLS-complete. To address this, we propose a fully polynomial time approximation scheme to identify an approximate PNE. Moreover, the tight Price of Anarchy (PoA) is established. In the asymmetric setting, we show the nonexistence of PNE in certain instances and establish that determining its existence is NP-hard. Nevertheless, we prove the existence of an approximate PNE. Additionally, when agents select topics sequentially, we demonstrate that finding a subgame-perfect equilibrium is PSPACE-hard. Furthermore, we present the sequential PoA for the two-agent setting.

cs.GT↗

arXiv2023

Competition among Pairwise Lottery Contests

We investigate a two-stage competitive model involving multiple contests. In this model, each contest designer chooses two participants from a pool of candidate contestants and determines the biases. Contestants strategically distribute their efforts across various contests within their budget. We first show the existence of a pure strategy Nash equilibrium (PNE) for the contestants, and propose a polynomial-time algorithm to compute an $ε$-approximate PNE. In the scenario where designers simultaneously decide the participants and biases, the subgame perfect equilibrium (SPE) may not exist. Nonetheless, when designers' decisions are made in two substages, the existence of SPE is established. In the scenario where designers can hold multiple contests, we show that the SPE exists under mild conditions and can be computed efficiently.

cs.GT↗

arXiv2023

TFLEX: Temporal Feature-Logic Embedding Framework for Complex Reasoning over Temporal Knowledge Graph

Multi-hop logical reasoning over knowledge graph (KG) plays a fundamental role in many artificial intelligence tasks. Recent complex query embedding (CQE) methods for reasoning focus on static KGs, while temporal knowledge graphs (TKGs) have not been fully explored. Reasoning over TKGs has two challenges: 1. The query should answer entities or timestamps; 2. The operators should consider both set logic on entity set and temporal logic on timestamp set. To bridge this gap, we define the multi-hop logical reasoning problem on TKGs. With generated three datasets, we propose the first temporal CQE named Temporal Feature-Logic Embedding framework (TFLEX) to answer the temporal complex queries. We utilize vector logic to compute the logic part of Temporal Feature-Logic embeddings, thus naturally modeling all First-Order Logic (FOL) operations on entity set. In addition, our framework extends vector logic on timestamp set to cope with three extra temporal operators (After, Before and Between). Experiments on numerous query patterns demonstrate the effectiveness of our method.

cs.LG↗

arXiv2023

On the Complexity of Computing Markov Perfect Equilibrium in General-Sum Stochastic Games

Similar to the role of Markov decision processes in reinforcement learning, Stochastic Games (SGs) lay the foundation for the study of multi-agent reinforcement learning (MARL) and sequential agent interactions. In this paper, we derive that computing an approximate Markov Perfect Equilibrium (MPE) in a finite-state discounted Stochastic Game within the exponential precision is \textbf{PPAD}-complete. We adopt a function with a polynomially bounded description in the strategy space to convert the MPE computation to a fixed-point problem, even though the stochastic game may demand an exponential number of pure strategies, in the number of states, for each agent. The completeness result follows the reduction of the fixed-point problem to {\sc End of the Line}. Our results indicate that finding an MPE in SGs is highly unlikely to be \textbf{NP}-hard unless \textbf{NP}=\textbf{co-NP}. Our work offers confidence for MARL research to study MPE computation on general-sum SGs and to develop fruitful algorithms as currently on zero-sum SGs.

cs.GT↗

arXiv2022

Competition among Parallel Contests

We investigate the model of multiple contests held in parallel, where each contestant selects one contest to join and each contest designer decides the prize structure to compete for the participation of contestants. We first analyze the strategic behaviors of contestants and completely characterize the symmetric Bayesian Nash equilibrium. As for the strategies of contest designers, when other designers' strategies are known, we show that computing the best response is NP-hard and propose a fully polynomial time approximation scheme (FPTAS) to output the $ε$-approximate best response. When other designers' strategies are unknown, we provide a worst case analysis on one designer's strategy. We give an upper bound on the utility of any strategy and propose a method to construct a strategy whose utility can guarantee a constant ratio of this upper bound in the worst case.

cs.GT↗

arXiv2022

On Convergence Lemma and Convergence Stability for Piecewise Analytic Functions

In this work, a convergence lemma for function $f$ being finite compositions of analytic mappings and the maximum operator is proved. The lemma shows that the set of $δ$-stationary points near an isolated local minimum point $x^*$ is shrinking to $x^*$ as $δ\to 0$. It is a natural extension of the version for strongly convex $C^1$ functions. However, the correctness of the lemma is subtle. Analytic mappings are necessary for the lemma in the sense that replacing it with differentiable or $C^\infty$ mappings makes the lemma false. The proof is based on stratification theorems of semi-analytic sets by Łojasiewicz. An extension of this proof presents a geometric characterization of the set of stationary points of $f$. Finally, a notion of stability on stationary points, called convergence stability, is proposed. It asks, under small numerical errors, whether a reasonable convergent optimization method started near a stationary point should eventually converge to the same stationary point. The concept of convergence stability becomes nontrivial qualitatively only when the objective function is both nonsmooth and nonconvex. Via the convergence lemma, an intuitive equivalent condition for convergence stability of $f$ is proved. These results together provide a new geometric perspective to study the problem of "where-to-converge" in nonsmooth nonconvex optimization.

cs.GT↗