When AI Generates Covariates: Causal Typing and Estimand Drift in Sequential Experiments
研究解决了AI生成协变量在顺序实验中导致因果问题定义变化的问题，通过提出一种包含版本表示映射、因果角色分类器等方法的因果类型学科来解决。
0 citationsRead paper
Selected work
Representative Papers
Information Set Emulation: Causal Certificates for AI Derived EHR Features
本文提出信息集仿真方法，通过AI提取电子健康记录特征并附带因果角色证据，以解决AI提取特征在因果推断中的适用性问题。
0 citationsRead paper
DTI-Guided Volumetric Spherical Harmonics Regression for Single-to-Multi-Shell dMRI Synthesis
本文提出DTI-SHNet，一种基于体积球谐函数回归的单壳至多壳dMRI合成方法，通过信号一致性正则化提高合成精度与保真度。
0 citationsRead paper
DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory
针对自进化过程中性能停滞或下降的问题，提出DiagEvo方法，通过分层错误记忆从解算器的失败历史中提取并存储错误原因，指导问题生成。
0 citationsRead paper
Energy Balancing Weights for Mediation Analysis
This study addresses the challenge of model dependence in reconstructing counterfactual distributions for causal mediation analysis by proposing the Energy Balance Weighting Method (EBWMA). By minimizing energy distance to directly approximate target joint distributions via sequential quadratic programming, EBWMA circumvents the need to model treatment mechanisms, mediator density ratios, or outcome regressions, thereby enabling model-free natural effect estimation. Experimental results demonstrate that under nonlinear and skewed data conditions, EBWMA significantly outperforms existing methods with lower bias and root mean squared error, reduced Monte Carlo variability, and optimal covariate balance in real-world applications.
0 citationsRead paper
Recent publications
Latest Papers
When AI Generates Covariates: Causal Typing and Estimand Drift in Sequential Experiments
研究解决了AI生成协变量在顺序实验中导致因果问题定义变化的问题，通过提出一种包含版本表示映射、因果角色分类器等方法的因果类型学科来解决。
0 citationsRead paper
Information Set Emulation: Causal Certificates for AI Derived EHR Features
本文提出信息集仿真方法，通过AI提取电子健康记录特征并附带因果角色证据，以解决AI提取特征在因果推断中的适用性问题。
0 citationsRead paper
DTI-Guided Volumetric Spherical Harmonics Regression for Single-to-Multi-Shell dMRI Synthesis
本文提出DTI-SHNet，一种基于体积球谐函数回归的单壳至多壳dMRI合成方法，通过信号一致性正则化提高合成精度与保真度。
0 citationsRead paper
DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory
针对自进化过程中性能停滞或下降的问题，提出DiagEvo方法，通过分层错误记忆从解算器的失败历史中提取并存储错误原因，指导问题生成。
0 citationsRead paper
Energy Balancing Weights for Mediation Analysis
This study addresses the challenge of model dependence in reconstructing counterfactual distributions for causal mediation analysis by proposing the Energy Balance Weighting Method (EBWMA). By minimizing energy distance to directly approximate target joint distributions via sequential quadratic programming, EBWMA circumvents the need to model treatment mechanisms, mediator density ratios, or outcome regressions, thereby enabling model-free natural effect estimation. Experimental results demonstrate that under nonlinear and skewed data conditions, EBWMA significantly outperforms existing methods with lower bias and root mean squared error, reduced Monte Carlo variability, and optimal covariate balance in real-world applications.
0 citationsRead paper