Selected work

Representative Papers

Hierarchical Group-Conditional Conformal Risk Control for Selective Prediction in Language Models

Jul 27, 2026

This work addresses the limitation of existing conformal risk control methods, which fail to enforce risk constraints across subgroups under shifting group structures, often leading to excessive error exposure for certain populations. To bridge the gap between marginal risk guarantees and group fairness, we propose HG-CRC, a novel framework that incorporates hierarchical group structure into conformal prediction. By simultaneously imposing risk constraints at all levels of the hierarchy and employing Bonferroni correction alongside a leaf-node-prioritized threshold selection strategy, HG-CRC enables fine-grained, post-hoc calibration without model retraining. Empirical evaluation on the ARC Challenge demonstrates that our method achieves a 0% empirical violation rate and WGER = 0 for high-accuracy models, while ablation studies confirm the critical role of hierarchy depth in satisfying the prescribed risk budget.

0 citationsRead paper

Income inequality estimation with gamma mixtures

Jul 06, 2026

This study addresses the accurate estimation of the m-th order Gini index under finite gamma mixture distributions with a common rate parameter. For the first time, closed-form expressions are derived for both the m-th order Gini index and its U-statistic estimator, and their theoretical properties—including strong consistency, asymptotic unbiasedness, and asymptotic normality—are rigorously established. The proposed methodology integrates bias correction, asymptotic analysis, and Monte Carlo simulation, demonstrating robust performance even when the common-rate assumption is relaxed. Empirical analyses confirm that the framework effectively captures inequality in real-world income distributions, offering a theoretically grounded and practically viable tool for higher-order inequality measurement.

0 citationsRead paper

Evaluating Large Language Models for Decision-Making in Agent-Based Urban Mobility Simulations

Jul 02, 2026

This study addresses the limited adaptability and contextual responsiveness of traditional rule-based multi-agent traffic simulations in dynamic environments. To overcome this, the authors propose a novel hybrid architecture that integrates a large language model (LLM) as a complementary cognitive layer within the GAMA simulation platform. The LLM is invoked via API to dynamically assess the need for route replanning and is augmented with a persistent memory mechanism to enhance behavioral consistency. While preserving conventional path-planning algorithms, this approach significantly improves agents’ situational awareness and decision-making flexibility. Experimental results demonstrate that, in highly dynamic scenarios, LLM-augmented agents exhibit superior adaptability and behavioral coherence, thereby validating the potential of LLMs for modeling spatially explicit multi-agent systems in urban traffic simulation.

0 citationsRead paper

A Novel Machine Learning Approach for Central Nervous System Tumor Classification from DNA Methylation

Jul 01, 2026

This study addresses the poor cross-cohort generalizability, methodological shortcomings, and unstable multi-class evaluation commonly observed in DNA methylation–based classification of central nervous system tumors. To overcome these limitations, we propose a streamlined machine learning pipeline integrating sparse random projection for dimensionality reduction with multinomial logistic regression, coupled with stratified cross-validation and a rigorous evaluation protocol. The approach achieves high generalization performance while maintaining model simplicity and clinical interpretability. Evaluated on a reference cohort of 2,801 samples, the method attains an average accuracy of 96%. In an independent clinical cohort of 1,104 cases, it yields 86% accuracy at the 91-class level and 93% at the family level—representing a 4–5 percentage point improvement over existing methods—and demonstrates strong practical utility for clinical deployment.

0 citationsRead paper

Unbiased estimation of normalized scale-invariant indices under the gamma distribution

Jun 21, 2026

This study addresses the lack of a general unbiased estimator for normalized scale-invariant indices—such as the Gini coefficient and entropy-based measures—under gamma-distributed populations. The authors construct such indices using homogeneous functions and leverage the independence between the sum of gamma variables and their Dirichlet-normalized proportions to develop unbiased estimators via U-statistics. This approach establishes, for the first time, a unified framework for unbiased estimation of any normalized scale-invariant index under gamma models. Theoretical analysis and Monte Carlo simulations demonstrate the robustness of the proposed estimators even under generalized gamma distributions. Empirical application to per capita GDP data across the Americas shows that the estimators perform consistently well across various indices and scenarios.

0 citationsRead paper

Recent publications

Latest Papers

Hierarchical Group-Conditional Conformal Risk Control for Selective Prediction in Language Models

Jul 27, 2026

This work addresses the limitation of existing conformal risk control methods, which fail to enforce risk constraints across subgroups under shifting group structures, often leading to excessive error exposure for certain populations. To bridge the gap between marginal risk guarantees and group fairness, we propose HG-CRC, a novel framework that incorporates hierarchical group structure into conformal prediction. By simultaneously imposing risk constraints at all levels of the hierarchy and employing Bonferroni correction alongside a leaf-node-prioritized threshold selection strategy, HG-CRC enables fine-grained, post-hoc calibration without model retraining. Empirical evaluation on the ARC Challenge demonstrates that our method achieves a 0% empirical violation rate and WGER = 0 for high-accuracy models, while ablation studies confirm the critical role of hierarchy depth in satisfying the prescribed risk budget.

0 citationsRead paper

Income inequality estimation with gamma mixtures

Jul 06, 2026

This study addresses the accurate estimation of the m-th order Gini index under finite gamma mixture distributions with a common rate parameter. For the first time, closed-form expressions are derived for both the m-th order Gini index and its U-statistic estimator, and their theoretical properties—including strong consistency, asymptotic unbiasedness, and asymptotic normality—are rigorously established. The proposed methodology integrates bias correction, asymptotic analysis, and Monte Carlo simulation, demonstrating robust performance even when the common-rate assumption is relaxed. Empirical analyses confirm that the framework effectively captures inequality in real-world income distributions, offering a theoretically grounded and practically viable tool for higher-order inequality measurement.

0 citationsRead paper

Evaluating Large Language Models for Decision-Making in Agent-Based Urban Mobility Simulations

Jul 02, 2026

This study addresses the limited adaptability and contextual responsiveness of traditional rule-based multi-agent traffic simulations in dynamic environments. To overcome this, the authors propose a novel hybrid architecture that integrates a large language model (LLM) as a complementary cognitive layer within the GAMA simulation platform. The LLM is invoked via API to dynamically assess the need for route replanning and is augmented with a persistent memory mechanism to enhance behavioral consistency. While preserving conventional path-planning algorithms, this approach significantly improves agents’ situational awareness and decision-making flexibility. Experimental results demonstrate that, in highly dynamic scenarios, LLM-augmented agents exhibit superior adaptability and behavioral coherence, thereby validating the potential of LLMs for modeling spatially explicit multi-agent systems in urban traffic simulation.

0 citationsRead paper

A Novel Machine Learning Approach for Central Nervous System Tumor Classification from DNA Methylation

Jul 01, 2026

This study addresses the poor cross-cohort generalizability, methodological shortcomings, and unstable multi-class evaluation commonly observed in DNA methylation–based classification of central nervous system tumors. To overcome these limitations, we propose a streamlined machine learning pipeline integrating sparse random projection for dimensionality reduction with multinomial logistic regression, coupled with stratified cross-validation and a rigorous evaluation protocol. The approach achieves high generalization performance while maintaining model simplicity and clinical interpretability. Evaluated on a reference cohort of 2,801 samples, the method attains an average accuracy of 96%. In an independent clinical cohort of 1,104 cases, it yields 86% accuracy at the 91-class level and 93% at the family level—representing a 4–5 percentage point improvement over existing methods—and demonstrates strong practical utility for clinical deployment.

0 citationsRead paper

Unbiased estimation of normalized scale-invariant indices under the gamma distribution

Jun 21, 2026

This study addresses the lack of a general unbiased estimator for normalized scale-invariant indices—such as the Gini coefficient and entropy-based measures—under gamma-distributed populations. The authors construct such indices using homogeneous functions and leverage the independence between the sum of gamma variables and their Dirichlet-normalized proportions to develop unbiased estimators via U-statistics. This approach establishes, for the first time, a unified framework for unbiased estimation of any normalized scale-invariant index under gamma models. Theoretical analysis and Monte Carlo simulations demonstrate the robustness of the proposed estimators even under generalized gamma distributions. Empirical application to per capita GDP data across the Americas shows that the estimators perform consistently well across various indices and scenarios.

0 citationsRead paper