Approved Too Late: Verdict Staleness in LLM-Guarded Self-Adaptive Systems
研究解决自适应系统中语言模型护栏判决过时问题，提出新鲜度边界防护方法估计批准的有效期，减少批准过期率。
0 citationsRead paper
Selected work
Representative Papers
Rethinking Higher Education: From Fixed Curricula to Learnity Graphs
Traditional higher education struggles to meet the demands of lifelong learning in the AI era, particularly regarding creativity, interdisciplinary integration, and practical experience. This work proposes a novel learning framework centered on the “Learnity Graph,” which leverages a structured graph model to interconnect academic, professional, and personal learning into unified knowledge, skill, and achievement units, thereby transcending rigid curricular boundaries. By integrating AI-driven knowledge organization and pathway recommendation mechanisms, the framework enables personalized, dynamic, and cross-domain learning trajectories. It fosters a scalable learning ecosystem that provides both theoretical grounding and practical pathways for transforming higher education toward lifelong learning.
0 citationsRead paper
The Complete Extended Euclidean Scheme Is Not in Piecewise Arithmetic $\mathrm{AC}^0$
This work investigates the computability of the extended Euclidean algorithm within the model of polynomial-size, constant-depth piecewise arithmetic circuits. By establishing, for the first time, an algebraic connection between Euclidean remainder sequences and Hankel determinants—and combining this with an analysis of principal subresultant coefficients, algebraic-geometric techniques over Zariski-open sets, and division elimination methods—the authors prove that the algorithm cannot be efficiently realized by such circuits. As corollaries, polynomial continued fraction expansions, bounded-degree principal subresultant coefficient profiles, and normalized subdiagonal Padé approximations also admit no polynomial-size, constant-depth piecewise arithmetic circuit implementations, thereby establishing strong circuit complexity lower bounds for several related algebraic computation problems.
0 citationsRead paper
The Degree of Strategy-Proofness for Risk-Averse Committee Selection
This study addresses the tension between proportional representation and strategic robustness in multi-winner approval voting, where the safety margin for risk-averse manipulation remains unclear. We extend the Risk-Averse Truthfulness (RAT) framework to approval-based committee elections and introduce, for the first time, the notion of RAT-degree. Focusing on Proportional Approval Voting (PAV), we combine techniques from mechanism design, combinatorial voting theory, and discrete mathematics to precisely characterize its vulnerability to safe coalition manipulation. Our analysis establishes tight bounds: PAV is susceptible to safe manipulation when manipulators possess at least ⌈n/k⌉ votes, yet remains fully immune if they control no more than ⌊n/(k+1)⌋ − 1 votes.
0 citationsRead paper
On Fréchet Traveling Salesmen Problems
This work proposes a novel variant of the Traveling Salesman Problem in which two agents must collaboratively traverse a given set of points, minimizing the Fréchet distance between their respective trajectories while simultaneously optimizing total path length and load balance. It introduces the Fréchet distance—commonly used to measure curve similarity—into multi-agent path planning for the first time, uncovering new applications in routing and network design. For the discrete Fréchet distance, the authors devise an efficient near-linear-time algorithm; in contrast, they rigorously establish the NP-hardness of the continuous case. By integrating techniques from computational geometry and combinatorial optimization, the study systematically analyzes multiple problem variants, carefully balancing theoretical complexity with practical tractability.
0 citationsRead paper
Recent publications
Latest Papers
Approved Too Late: Verdict Staleness in LLM-Guarded Self-Adaptive Systems
研究解决自适应系统中语言模型护栏判决过时问题，提出新鲜度边界防护方法估计批准的有效期，减少批准过期率。
0 citationsRead paper
Rethinking Higher Education: From Fixed Curricula to Learnity Graphs
Traditional higher education struggles to meet the demands of lifelong learning in the AI era, particularly regarding creativity, interdisciplinary integration, and practical experience. This work proposes a novel learning framework centered on the “Learnity Graph,” which leverages a structured graph model to interconnect academic, professional, and personal learning into unified knowledge, skill, and achievement units, thereby transcending rigid curricular boundaries. By integrating AI-driven knowledge organization and pathway recommendation mechanisms, the framework enables personalized, dynamic, and cross-domain learning trajectories. It fosters a scalable learning ecosystem that provides both theoretical grounding and practical pathways for transforming higher education toward lifelong learning.
0 citationsRead paper
The Complete Extended Euclidean Scheme Is Not in Piecewise Arithmetic $\mathrm{AC}^0$
This work investigates the computability of the extended Euclidean algorithm within the model of polynomial-size, constant-depth piecewise arithmetic circuits. By establishing, for the first time, an algebraic connection between Euclidean remainder sequences and Hankel determinants—and combining this with an analysis of principal subresultant coefficients, algebraic-geometric techniques over Zariski-open sets, and division elimination methods—the authors prove that the algorithm cannot be efficiently realized by such circuits. As corollaries, polynomial continued fraction expansions, bounded-degree principal subresultant coefficient profiles, and normalized subdiagonal Padé approximations also admit no polynomial-size, constant-depth piecewise arithmetic circuit implementations, thereby establishing strong circuit complexity lower bounds for several related algebraic computation problems.
0 citationsRead paper
The Degree of Strategy-Proofness for Risk-Averse Committee Selection
This study addresses the tension between proportional representation and strategic robustness in multi-winner approval voting, where the safety margin for risk-averse manipulation remains unclear. We extend the Risk-Averse Truthfulness (RAT) framework to approval-based committee elections and introduce, for the first time, the notion of RAT-degree. Focusing on Proportional Approval Voting (PAV), we combine techniques from mechanism design, combinatorial voting theory, and discrete mathematics to precisely characterize its vulnerability to safe coalition manipulation. Our analysis establishes tight bounds: PAV is susceptible to safe manipulation when manipulators possess at least ⌈n/k⌉ votes, yet remains fully immune if they control no more than ⌊n/(k+1)⌋ − 1 votes.
0 citationsRead paper
On Fréchet Traveling Salesmen Problems
This work proposes a novel variant of the Traveling Salesman Problem in which two agents must collaboratively traverse a given set of points, minimizing the Fréchet distance between their respective trajectories while simultaneously optimizing total path length and load balance. It introduces the Fréchet distance—commonly used to measure curve similarity—into multi-agent path planning for the first time, uncovering new applications in routing and network design. For the discrete Fréchet distance, the authors devise an efficient near-linear-time algorithm; in contrast, they rigorously establish the NP-hardness of the continuous case. By integrating techniques from computational geometry and combinatorial optimization, the study systematically analyzes multiple problem variants, carefully balancing theoretical complexity with practical tractability.
0 citationsRead paper