Slow Decay and Silenced Expression: Iterated Subliminal Trait Transfer in Language-Model Lineages
研究通过迭代训练方法探讨语言模型中潜意识特征在多代传递中的持久性与衰减，使用关键词筛选和激活探针评估特征表达。
0 citationsRead paper
Selected work
Representative Papers
RKHS-Based Inference for Nonlinear Granger Causality via Conditional Centering
本文提出了一种基于再生核希尔伯特空间（RKHS）的方法，用于检测非线性自回归过程中的非线性格兰杰因果关系，解决了传统方法只能检测线性预测关系的问题。
0 citationsRead paper
The Cost of Failure: On The Complexity of Recampaigning under Fixed Districts
This study investigates whether a disadvantaged political party can strategically reallocate its candidates across fixed electoral districts to secure more seats. It formally introduces and defines the “re-contesting” problem for the first time and provides a systematic analysis of its computational complexity. Employing polynomial-time many-one reductions, parameterized complexity theory, and axiomatic methods, the work delineates precise boundaries between tractable and intractable cases: it characterizes conditions under which the problem is solvable in polynomial time versus NP-hard, and establishes reduction and separation relationships among multiple modeling variants. These results offer a rigorous theoretical foundation for understanding the computational feasibility of strategic electoral interventions under fixed district boundaries.
0 citationsRead paper
Carrying is Hard: Exploring the Gap between Hardness for NP and PSPACE for the Hanano and Jelly no Puzzles
This study investigates why Hanano Puzzle remains PSPACE-complete under the same rule constraints that render Jelly no Puzzle merely NP-complete. Through computational complexity theory, reduction proofs, and formal modeling of game mechanics, the work identifies and rigorously defines the “carrying” mechanism—whereby blocks can be moved indirectly by being carried by other blocks—as the key factor responsible for the PSPACE-hardness of Hanano Puzzle. This insight not only clarifies the fundamental source of the complexity gap between these two puzzle types but also contributes to a deeper understanding of the essential distinctions between PSPACE and NP problems.
0 citationsRead paper
Approximating Electoral Control Problems
This paper investigates the approximability of election control problems under plurality, approval, and Condorcet voting rules, establishing tight approximation ratios for standard control problems in both weighted and unweighted voter settings. Methodologically, it introduces the Minimum k-Union problem to computational social choice and develops the first generic $O(log n)$-approximation framework based on covering integer programming. For plurality, it presents an $O(m)$-approximation algorithm and proves a matching $Omega(m^{1/4})$ lower bound—tight up to polynomial factors. Via axiomatic generalization, the results extend to infinite families of voting rules. The work resolves multiple long-standing open problems posed over the past 18 years, fully characterizing the optimal or asymptotically optimal approximability of 12 distinct control problem variants. Collectively, these contributions substantially advance the theoretical frontier of approximation algorithms for election control.
0 citationsRead paper
Recent publications
Latest Papers
Slow Decay and Silenced Expression: Iterated Subliminal Trait Transfer in Language-Model Lineages
研究通过迭代训练方法探讨语言模型中潜意识特征在多代传递中的持久性与衰减，使用关键词筛选和激活探针评估特征表达。
0 citationsRead paper
RKHS-Based Inference for Nonlinear Granger Causality via Conditional Centering
本文提出了一种基于再生核希尔伯特空间（RKHS）的方法，用于检测非线性自回归过程中的非线性格兰杰因果关系，解决了传统方法只能检测线性预测关系的问题。
0 citationsRead paper
The Cost of Failure: On The Complexity of Recampaigning under Fixed Districts
This study investigates whether a disadvantaged political party can strategically reallocate its candidates across fixed electoral districts to secure more seats. It formally introduces and defines the “re-contesting” problem for the first time and provides a systematic analysis of its computational complexity. Employing polynomial-time many-one reductions, parameterized complexity theory, and axiomatic methods, the work delineates precise boundaries between tractable and intractable cases: it characterizes conditions under which the problem is solvable in polynomial time versus NP-hard, and establishes reduction and separation relationships among multiple modeling variants. These results offer a rigorous theoretical foundation for understanding the computational feasibility of strategic electoral interventions under fixed district boundaries.
0 citationsRead paper
Carrying is Hard: Exploring the Gap between Hardness for NP and PSPACE for the Hanano and Jelly no Puzzles
This study investigates why Hanano Puzzle remains PSPACE-complete under the same rule constraints that render Jelly no Puzzle merely NP-complete. Through computational complexity theory, reduction proofs, and formal modeling of game mechanics, the work identifies and rigorously defines the “carrying” mechanism—whereby blocks can be moved indirectly by being carried by other blocks—as the key factor responsible for the PSPACE-hardness of Hanano Puzzle. This insight not only clarifies the fundamental source of the complexity gap between these two puzzle types but also contributes to a deeper understanding of the essential distinctions between PSPACE and NP problems.
0 citationsRead paper
Approximating Electoral Control Problems
This paper investigates the approximability of election control problems under plurality, approval, and Condorcet voting rules, establishing tight approximation ratios for standard control problems in both weighted and unweighted voter settings. Methodologically, it introduces the Minimum k-Union problem to computational social choice and develops the first generic $O(log n)$-approximation framework based on covering integer programming. For plurality, it presents an $O(m)$-approximation algorithm and proves a matching $Omega(m^{1/4})$ lower bound—tight up to polynomial factors. Via axiomatic generalization, the results extend to infinite families of voting rules. The work resolves multiple long-standing open problems posed over the past 18 years, fully characterizing the optimal or asymptotically optimal approximability of 12 distinct control problem variants. Collectively, these contributions substantially advance the theoretical frontier of approximation algorithms for election control.
0 citationsRead paper