Draining Fictitious Knots: Restoring Distance-Awareness Guarantees for High-Dimensional Spline Networks
研究解决了高维KANs中虚构结点导致距离感知保证失效的问题，通过引入排水不确定性机制恢复距离感知。
0 citationsRead paper
Selected work
Representative Papers
SoK: From Generation to Consumption of Privacy Documents in Software Systems
This study addresses the long-standing lack of a unified software engineering perspective on privacy documents throughout their lifecycle, which has led to fragmented approaches in generation, analysis, compliance verification, and usability evaluation. Through a systematic literature review (SoK) of 290 studies published between 2010 and 2025, this work proposes the first comprehensive lifecycle framework encompassing definition, generation, analysis, compliance validation, and usability assessment. The research identifies 15 key trends, 21 open challenges, and four major future directions, with particular emphasis on leveraging large language models for analyzing consistency between privacy policies and code implementations. By establishing a structured knowledge base and introducing a novel paradigm that balances usability for both end users and developers, this work lays a shared foundation for privacy documentation research in the AI era.
0 citationsRead paper
A Computational Model for Measuring Adaptability Among U.S. Farmers: Evidence from 1997-2022
This study investigates how U.S. farmers’ crop choices reflect environmental adaptation and drive the evolution of regional agro-cultural traits. Leveraging county-level agricultural panel data from 1997 to 2022, it pioneers an integration of cultural evolution theory with empirical agricultural data within a computational social science framework to quantify how environmental returns shape crop portfolio selection. The analysis reveals that counties consistently adopt crop combinations that maximize local environmental fit and productivity, exhibiting clear long-term adaptive trajectories. These findings uncover a mechanism of cumulative cultural selection driven by environmental returns, offering novel evidence for understanding the evolutionary logic of human behavior in ecological adaptation.
0 citationsRead paper
HARP: Measuring Harm Amplification in Multi-Agent LLM Systems
This work addresses the critical yet underexplored risk that localized perturbations in multi-agent large language model systems can be amplified through inter-agent interactions into system-wide harms, a phenomenon inadequately quantified by existing evaluation methods. To bridge this gap, the authors propose HARP, a novel framework that introduces the first execution-trajectory-based metric for measuring local-to-global harm amplification. By comparing clean and perturbed execution trajectories, HARP tracks behavioral divergences in outputs, tool invocations, and memory operations to compute a harm amplification ratio. The study implements targeted attack mechanisms—including role-based perturbation injection, shared context manipulation, and temporal/memory persistence attacks—and introduces IntegrityGuard, a new defense strategy. Experiments on a seven-agent financial system demonstrate that single-point compromises can trigger maximal harm amplification, with shared context poisoning being the most effective attack vector; IntegrityGuard substantially mitigates global harm, albeit with trade-offs between utility and computational cost.
0 citationsRead paper
Knowledge Graph Representations for LLM-Based Policy Compliance Reasoning
This work addresses the compliance challenges posed by the rapid integration of artificial intelligence by proposing a knowledge graph–enhanced large language model (LLM) reasoning framework that efficiently and accurately extracts and reasons over compliance information from complex policy documents. The approach employs an agent-based architecture to automatically construct policy knowledge graphs and compares manually defined ontologies with those automatically discovered by LLMs, revealing the latter’s superior performance in cross-policy reasoning tasks. Comprehensive experiments across six question-answering categories, five LLMs, and multiple evaluation strategies consistently demonstrate—across 42 tasks—that integrating knowledge graphs significantly enhances LLMs’ compliance reasoning capabilities while exhibiting strong generalization.
0 citationsRead paper
Recent publications
Latest Papers
Draining Fictitious Knots: Restoring Distance-Awareness Guarantees for High-Dimensional Spline Networks
研究解决了高维KANs中虚构结点导致距离感知保证失效的问题，通过引入排水不确定性机制恢复距离感知。
0 citationsRead paper
SoK: From Generation to Consumption of Privacy Documents in Software Systems
This study addresses the long-standing lack of a unified software engineering perspective on privacy documents throughout their lifecycle, which has led to fragmented approaches in generation, analysis, compliance verification, and usability evaluation. Through a systematic literature review (SoK) of 290 studies published between 2010 and 2025, this work proposes the first comprehensive lifecycle framework encompassing definition, generation, analysis, compliance validation, and usability assessment. The research identifies 15 key trends, 21 open challenges, and four major future directions, with particular emphasis on leveraging large language models for analyzing consistency between privacy policies and code implementations. By establishing a structured knowledge base and introducing a novel paradigm that balances usability for both end users and developers, this work lays a shared foundation for privacy documentation research in the AI era.
0 citationsRead paper
A Computational Model for Measuring Adaptability Among U.S. Farmers: Evidence from 1997-2022
This study investigates how U.S. farmers’ crop choices reflect environmental adaptation and drive the evolution of regional agro-cultural traits. Leveraging county-level agricultural panel data from 1997 to 2022, it pioneers an integration of cultural evolution theory with empirical agricultural data within a computational social science framework to quantify how environmental returns shape crop portfolio selection. The analysis reveals that counties consistently adopt crop combinations that maximize local environmental fit and productivity, exhibiting clear long-term adaptive trajectories. These findings uncover a mechanism of cumulative cultural selection driven by environmental returns, offering novel evidence for understanding the evolutionary logic of human behavior in ecological adaptation.
0 citationsRead paper
HARP: Measuring Harm Amplification in Multi-Agent LLM Systems
This work addresses the critical yet underexplored risk that localized perturbations in multi-agent large language model systems can be amplified through inter-agent interactions into system-wide harms, a phenomenon inadequately quantified by existing evaluation methods. To bridge this gap, the authors propose HARP, a novel framework that introduces the first execution-trajectory-based metric for measuring local-to-global harm amplification. By comparing clean and perturbed execution trajectories, HARP tracks behavioral divergences in outputs, tool invocations, and memory operations to compute a harm amplification ratio. The study implements targeted attack mechanisms—including role-based perturbation injection, shared context manipulation, and temporal/memory persistence attacks—and introduces IntegrityGuard, a new defense strategy. Experiments on a seven-agent financial system demonstrate that single-point compromises can trigger maximal harm amplification, with shared context poisoning being the most effective attack vector; IntegrityGuard substantially mitigates global harm, albeit with trade-offs between utility and computational cost.
0 citationsRead paper
Knowledge Graph Representations for LLM-Based Policy Compliance Reasoning
This work addresses the compliance challenges posed by the rapid integration of artificial intelligence by proposing a knowledge graph–enhanced large language model (LLM) reasoning framework that efficiently and accurately extracts and reasons over compliance information from complex policy documents. The approach employs an agent-based architecture to automatically construct policy knowledge graphs and compares manually defined ontologies with those automatically discovered by LLMs, revealing the latter’s superior performance in cross-policy reasoning tasks. Comprehensive experiments across six question-answering categories, five LLMs, and multiple evaluation strategies consistently demonstrate—across 42 tasks—that integrating knowledge graphs significantly enhances LLMs’ compliance reasoning capabilities while exhibiting strong generalization.
0 citationsRead paper