Distortion of AI Alignment Revisited: RLHF is a Decent Utilitarian Aligner
本文分析了RLHF在用户偏好不同时效用失真的问题，通过奖励裁剪方法改善了因分布不匹配导致的效用失真，并提出了优化建议。
0 citationsRead paper
Selected work
Representative Papers
RENSA: Rich Environment Metadata to Navigate Shared and Distributed Endpoints for Automated Federated SPARQL Query Generation
为解决联邦环境下构建高效SPARQL查询的难题，提出RENSA框架，通过扩展SBM元数据来实现精确的数据源选择和语义约束推断，减少通信开销。
0 citationsRead paper
Frequentist prediction intervals for random-effects meta-analysis via confidence-distribution propagation
该研究针对随机效应荟萃分析中预测区间覆盖率不足的问题，提出了一种基于置信分布传播的方法来改进预测区间的准确性。
0 citationsRead paper
One-Sided-Error Parameterized Reductions for the Minimum Distance and Shortest Vector Problems
This study addresses the longstanding challenge of establishing deterministic reductions and parameterized approximation hardness for the Minimum Distance Problem (MDP) and Shortest Vector Problem (SVP). We propose a one-sided error randomized reduction framework that leverages OR-function constructions and achieves conditional derandomization under circuit lower bound assumptions. Our results establish both NP-hardness and W[1]-hardness for MDP and SVP under constant-factor approximations. These findings bridge critical gaps in deterministic and parameterized complexity theory, providing novel theoretical foundations for understanding the computational nature of these fundamental lattice problems.
0 citationsRead paper
On the Complexity of Locally Dense Lattices
This study addresses the decision complexity and definitional equivalence of locally dense lattices by introducing the LDL P decision problem and analyzing it through the polynomial hierarchy. We prove that this problem is Σ₂^P-complete under the infinity norm when p ≥ log₂3. Furthermore, we establish the equivalence between two prevailing definitions via deterministic polynomial-time reductions. By precisely characterizing the computational complexity class of locally dense lattice recognition and unifying classical definitions, this work provides a rigorous theoretical foundation for future research in lattice-based complexity theory and related domains.
0 citationsRead paper
Recent publications
Latest Papers
Distortion of AI Alignment Revisited: RLHF is a Decent Utilitarian Aligner
本文分析了RLHF在用户偏好不同时效用失真的问题，通过奖励裁剪方法改善了因分布不匹配导致的效用失真，并提出了优化建议。
0 citationsRead paper
RENSA: Rich Environment Metadata to Navigate Shared and Distributed Endpoints for Automated Federated SPARQL Query Generation
为解决联邦环境下构建高效SPARQL查询的难题，提出RENSA框架，通过扩展SBM元数据来实现精确的数据源选择和语义约束推断，减少通信开销。
0 citationsRead paper
Frequentist prediction intervals for random-effects meta-analysis via confidence-distribution propagation
该研究针对随机效应荟萃分析中预测区间覆盖率不足的问题，提出了一种基于置信分布传播的方法来改进预测区间的准确性。
0 citationsRead paper
One-Sided-Error Parameterized Reductions for the Minimum Distance and Shortest Vector Problems
This study addresses the longstanding challenge of establishing deterministic reductions and parameterized approximation hardness for the Minimum Distance Problem (MDP) and Shortest Vector Problem (SVP). We propose a one-sided error randomized reduction framework that leverages OR-function constructions and achieves conditional derandomization under circuit lower bound assumptions. Our results establish both NP-hardness and W[1]-hardness for MDP and SVP under constant-factor approximations. These findings bridge critical gaps in deterministic and parameterized complexity theory, providing novel theoretical foundations for understanding the computational nature of these fundamental lattice problems.
0 citationsRead paper
On the Complexity of Locally Dense Lattices
This study addresses the decision complexity and definitional equivalence of locally dense lattices by introducing the LDL P decision problem and analyzing it through the polynomial hierarchy. We prove that this problem is Σ₂^P-complete under the infinity norm when p ≥ log₂3. Furthermore, we establish the equivalence between two prevailing definitions via deterministic polynomial-time reductions. By precisely characterizing the computational complexity class of locally dense lattice recognition and unifying classical definitions, this work provides a rigorous theoretical foundation for future research in lattice-based complexity theory and related domains.
0 citationsRead paper