The Ethics of Artificial Intelligence in Military Operations
本文探讨了军事行动中AI系统的伦理问题，特别是责任归属难题，并提出了一套包含明确责任角色、对抗性审计等措施的治理框架来解决这个问题。
0 citationsRead paper
Selected work
Representative Papers
On the Computation of Normalized Power Priors
Standardized power priors are hindered in widespread Bayesian applications due to the intractability of their normalizing constants. This work proposes a general Markov chain Monte Carlo (MCMC) strategy that circumvents explicit computation of the marginal likelihood by exploiting a functional identity between the marginal distribution of the power parameter and the normalizing constant. The approach integrates seamlessly into mainstream Bayesian software platforms such as Stan, JAGS, and NIMBLE, substantially lowering the computational barrier to implementation. Empirical evaluations across multiple case studies demonstrate that the method is both practical and efficient, offering a viable pathway for routine use of standardized power priors in Bayesian inference.
0 citationsRead paper
Stochastic Claims Reserving Using State Space Modeling
This paper addresses the challenge of estimating incurred-but-not-reported (IBNR) claim reserves in insurance. Methodologically, it proposes a stochastic reserving framework based on state-space models (SSMs), and—uniquely—integrates Kalman filtering, Kalman smoothing, and simulation smoothing into a unified end-to-end inference pipeline for IBNR prediction, model diagnostics, and sampling-based reserve distribution estimation. Key contributions include: (i) establishing a transparent, auditable, and traceable reserving workflow; (ii) providing a general-purpose SSM specification paradigm and principled model selection guidance; and (iii) releasing fully reproducible, open-source code implementing the framework within SAS Viya (via the CSSM procedure). Empirical evaluation on real-world insurance data demonstrates that the framework robustly quantifies both point estimates and uncertainty of IBNR reserves, balancing theoretical rigor with practical implementability.
0 citationsRead paper
Semiparametric Growth-Curve Modeling in Hierarchical, Longitudinal Studies
Comparing group-level growth (or decay) curves in hierarchical longitudinal data remains challenging: conventional parametric models (e.g., logistic, Gompertz) often fail to capture non-ideal monotonic trajectories and struggle to balance prior structural assumptions with data fidelity. To address this, we propose a semi-parametric functional mixed-effects state-space model. Our approach integrates parametric prior constraints with Bayesian nonparametric smoothing within a unified state-space framework, enabling simultaneous modeling of individual heterogeneity and inference of group-level functional trajectories. Compared to existing methods, it preserves interpretability while substantially improving fit accuracy and statistical power. Empirical evaluation on real bacterial growth data demonstrates that the model markedly enhances individual-level dynamic calibration and significantly increases the statistical power for detecting inter-group differences.
0 citationsRead paper
Recent publications
Latest Papers
The Ethics of Artificial Intelligence in Military Operations
本文探讨了军事行动中AI系统的伦理问题，特别是责任归属难题，并提出了一套包含明确责任角色、对抗性审计等措施的治理框架来解决这个问题。
0 citationsRead paper
On the Computation of Normalized Power Priors
Standardized power priors are hindered in widespread Bayesian applications due to the intractability of their normalizing constants. This work proposes a general Markov chain Monte Carlo (MCMC) strategy that circumvents explicit computation of the marginal likelihood by exploiting a functional identity between the marginal distribution of the power parameter and the normalizing constant. The approach integrates seamlessly into mainstream Bayesian software platforms such as Stan, JAGS, and NIMBLE, substantially lowering the computational barrier to implementation. Empirical evaluations across multiple case studies demonstrate that the method is both practical and efficient, offering a viable pathway for routine use of standardized power priors in Bayesian inference.
0 citationsRead paper
Stochastic Claims Reserving Using State Space Modeling
This paper addresses the challenge of estimating incurred-but-not-reported (IBNR) claim reserves in insurance. Methodologically, it proposes a stochastic reserving framework based on state-space models (SSMs), and—uniquely—integrates Kalman filtering, Kalman smoothing, and simulation smoothing into a unified end-to-end inference pipeline for IBNR prediction, model diagnostics, and sampling-based reserve distribution estimation. Key contributions include: (i) establishing a transparent, auditable, and traceable reserving workflow; (ii) providing a general-purpose SSM specification paradigm and principled model selection guidance; and (iii) releasing fully reproducible, open-source code implementing the framework within SAS Viya (via the CSSM procedure). Empirical evaluation on real-world insurance data demonstrates that the framework robustly quantifies both point estimates and uncertainty of IBNR reserves, balancing theoretical rigor with practical implementability.
0 citationsRead paper
Semiparametric Growth-Curve Modeling in Hierarchical, Longitudinal Studies
Comparing group-level growth (or decay) curves in hierarchical longitudinal data remains challenging: conventional parametric models (e.g., logistic, Gompertz) often fail to capture non-ideal monotonic trajectories and struggle to balance prior structural assumptions with data fidelity. To address this, we propose a semi-parametric functional mixed-effects state-space model. Our approach integrates parametric prior constraints with Bayesian nonparametric smoothing within a unified state-space framework, enabling simultaneous modeling of individual heterogeneity and inference of group-level functional trajectories. Compared to existing methods, it preserves interpretability while substantially improving fit accuracy and statistical power. Empirical evaluation on real bacterial growth data demonstrates that the model markedly enhances individual-level dynamic calibration and significantly increases the statistical power for detecting inter-group differences.
0 citationsRead paper