arXiv subjects

Zihan Luo

Publications and source records attributed to Zihan Luo.

7 recordsLinked to original sources

arXiv2026

Behavioral Grammar: Detecting Adaptive Malware via Tiny Language Model Priors and Second-Order Temporal Analysis

Modern endpoint detection systems face a fundamental tension: signature-based approaches are trivially evaded by polymorphic or adaptive threats, while heavy deep-learning models resist auditability and deployment at scale. This paper presents Behavioral Grammar, a detection architecture that treats host runtime behavior as a structured language and learns its "grammar" with a compact 0.88M-parameter causal Transformer (TinyGPT). Each system event is discretized into an 8-token representation spanning event type, process, argument skeleton, path category, parent process, user, destination, and inter-event timing. The model learns the conditional distribution of normal behavior in a purely self-supervised manner, and anomaly scores are derived from per-slot negative log-likelihood (NLL) statistics, yielding a mathematically bounded false-positive rate. We augment this prior with prototype learning for known-attack attribution, second-order temporal analysis for cadence-based detection, self-learning pattern extraction, and a five-network fusion pipeline. Against an Adaptive Adversarial Agent (AAA)--a threat that learns survival strategies under defensive pressure, performs behavioral mimicry, employs indirect execution via shell, and matches host event rates--our system achieves 93% detection at a 3.84% onboarding false-positive rate. The strongest discriminative signal arises not from any single event but from the coefficient of variation of inter-event intervals: the AAA stepping cadence exhibits CV = 0.310 versus 9.786 for benign sleep intervals, a 30x separation reflecting a fundamental stealth-functionality trade-off. We frame these findings within a coevolutionary economics model, arguing that behavior-grammar detection shifts the evasion cost from rule circumvention (cheap) to distribution matching (expensive), establishing a structural asymmetry that favors the defender.

cs.AI↗

arXiv2026

ToolRec: Calibrated Preference Alignment for Query Recommendation in On-Device Assistants

Large Language Models (LLMs) have significantly advanced generative query recommendation. However, while alignment is crucial for tailoring LLMs to human preferences, existing alignment methods primarily focus on standard chatbot scenarios, falling short in on-device intelligent assistants where users predominantly expect the rapid invocation of system-level tools. Moreover, directly aligning LLMs with real-world click logs introduces severe noise due to varying user activity levels and the failure to emphasize execution-oriented queries. To address these challenges, we propose ToolRec, a calibrated preference alignment framework tailored for on-device query recommendation. To ground query recommendation with executable tools, we first construct SysToolKit, a comprehensive repository of 708 system tools, paired with a context-aware tool retrieval mechanism to ensure that the extracted tools closely match the user's intent. A dual-level calibration mechanism is then proposed to refine raw click data, effectively mitigating user behavioral noise by calibrating signals based on user activity (user-level), while simultaneously up-weighting click signals on tool-invoking queries (system-level). Guided by these refined preference signals, we then align the model using a sample-level weighted Kahneman-Tversky Optimization (KTO). Extensive online A/B tests on our mobile assistant platform OPPO Xiaobu, which has over 150 million monthly active users, demonstrate that ToolRec can significantly improve Click-Through Rate (CTR) and total click volume over strong baselines while maintaining high query relevance.

cs.IR↗

arXiv2026

Strengthening Bulow-Klemperer-Style Results for Multi-Unit Auctions

The classic result of Bulow and Klemperer (1996) shows that in multi-unit auctions with $m$ units and $n\geq m$ buyers whose values are sampled i.i.d. from a regular distribution, the revenue of the VCG auction with $m$ additional buyers is at least as large as the optimal revenue. Unfortunately, for regular distributions, adding $m$ additional buyers is sometimes indeed necessary, so the "competition complexity" of the VCG auction is $m$. We seek proving better competition complexity results in two dimensions. First, under stronger distributional assumptions, the competition complexity of VCG auction drops dramatically. In balanced markets (where $m=n$) with MHR distributions, it is sufficient to only add $(e^{1/e} - 1 + o(1))n \approx 0.4447n$ additional buyers to match the optimal revenue -- less than half the number that is necessary under regularity -- and this bound is asymptotically tight. We provide both exact finite-market results for small value of $n$, and closed-form asymptotic formulas for general market with any $m\leq n$, and any target fraction of the optimal revenue. Second, we analyze a supply-limiting variant of VCG auction that caps the number of units sold in a prior-independent way. Whenever the goal is to achieve almost the optimal revenue, this mechanism strictly improves upon standard VCG auction, requiring significantly fewer additional buyers. Together, our results show that both stronger distributional assumptions, as well as a simple prior-independent refinement to the VCG auction, can each substantially reduce the number of additional buyers that is sufficient to achieve (near-)optimal revenue. Our analysis hinges on a unified worst-case reduction to truncated generalized Pareto distributions, enabling both numerical computation and analytical tractability.

cs.GT↗

arXiv2026

On the Fair Allocation to Asymmetric Agents with Binary XOS Valuations

We study the problem of allocating $m$ indivisible goods among $n$ agents, where each agent's valuation is fractionally subadditive (XOS). With respect to AnyPrice Share (APS) fairness, Kulkarni et al. (2024) showed that, when agents have binary marginal values, a $0.1222$-APS allocation can be found in polynomial time, and there exists an instance where no allocation is better than $0.5$-approximate APS. Very recently, Feige and Grinberg (2025) extended the problem to the asymmetric case, where agents may have different entitlements, and improved the approximation ratio to $1/6$ for general XOS valuations. In this work, we focus on the asymmetric setting with binary XOS valuations, and further improve the approximation ratio to $1/2$, which matches the known upper bound. We also present a polynomial-time algorithm to compute such an allocation. Beyond APS fairness, we also study the weighted maximin share (WMMS) fairness. Farhadi et al. (2019) showed that, a $1/n$-WMMS allocation always exists for agents with general additive valuations, and that this approximation ratio is tight. We extend this result to general XOS valuations, where a $1/n$-WMMS allocation still exists, and this approximation ratio cannot be improved even when marginal values are binary. This shows a sharp contrast to binary additive valuations, where an exact WMMS allocation exists and can be found in polynomial time.

cs.GT↗

arXiv2024

Are Your Models Still Fair? Fairness Attacks on Graph Neural Networks via Node Injections

Despite the remarkable capabilities demonstrated by Graph Neural Networks (GNNs) in graph-related tasks, recent research has revealed the fairness vulnerabilities in GNNs when facing malicious adversarial attacks. However, all existing fairness attacks require manipulating the connectivity between existing nodes, which may be prohibited in reality. To this end, we introduce a Node Injection-based Fairness Attack (NIFA), exploring the vulnerabilities of GNN fairness in such a more realistic setting. In detail, NIFA first designs two insightful principles for node injection operations, namely the uncertainty-maximization principle and homophily-increase principle, and then optimizes injected nodes' feature matrix to further ensure the effectiveness of fairness attacks. Comprehensive experiments on three real-world datasets consistently demonstrate that NIFA can significantly undermine the fairness of mainstream GNNs, even including fairness-aware GNNs, by injecting merely 1% of nodes. We sincerely hope that our work can stimulate increasing attention from researchers on the vulnerability of GNN fairness, and encourage the development of corresponding defense mechanisms. Our code and data are released at: https://github.com/CGCL-codes/NIFA.

cs.LG↗

arXiv2024

Design and Characterization of Strategy-Proof Mechanisms for Two-Facility Game on a Line

We focus on the problem of placing two facilities along a linear space to serve a group of agents. Each agent is committed to minimizing the distance between her location and the closest facility. A mechanism is an algorithm that maps the reported agent locations to the facility locations. We are interested in mechanisms without money that are deterministic, strategy-proof, and provide a bounded approximation ratio for social cost. It is a fundamental problem to characterize the family of strategy-proof mechanisms with a bounded approximation ratio. Fotakis and Tzamos already demonstrated that the deterministic strategy-proof mechanisms for the 2-facility game problem are mechanisms with a unique dictator and the leftmost-rightmost mechanism. In this paper, we first present a more refined characterization of the first family. We then reveal three new classes of strategy-proof mechanisms that show the intricacy of structure within this family. This helps us get a more complete picture of the characterization of the 2-facility game problem, and may also have value in understanding and solving more general facility allocation game problems. Besides, based on our refined characterization, we surprisingly find that prediction cannot effectively improve the performance of the mechanism in the two-facility game problem, while this methodology to overcome bad approximation ratio works in many other mechanism design problems. We show that if we require that the mechanism admits a bounded approximation ratio when the prediction is arbitrarily bad, then at the same time, the mechanism can never achieve sublinear approximation ratios even with perfect prediction.

cs.GT↗

arXiv2024

GraphInstruct: Empowering Large Language Models with Graph Understanding and Reasoning Capability

Improving the general capabilities of large language models (LLMs) is an active research topic. As a common data structure in many real-world domains, understanding graph data is a crucial part of advancing general intelligence. To this end, we propose a dynamic benchmark named GraphInstruct in this paper, which comprehensively includes 21 classical graph reasoning tasks, providing diverse graph generation pipelines and detailed intermediate reasoning steps for each sample. Based on GraphInstruct, we develop GraphSolver via efficient instruction-tuning, which demonstrates prominent graph understanding capability compared to other open-sourced LLMs. To further endow LLMs with multi-step graph reasoning capability, we propose a label-mask training strategy and build GraphSolver+, which leverages masked supervision on intermediate reasoning tokens to emphasize crucial node-identification signals. As one of the pioneering efforts to enhance the graph understanding and reasoning abilities of LLMs, extensive experiments have demonstrated the superiority of GraphSolver and GraphSolver+ over other LLMs. We sincerely hope GraphInstruct will facilitate further research on applying LLMs to graph-structured data. Our code and data are released publicly at: https://github.com/CGCL-codes/GraphInstruct.

cs.AI↗