Selected work

Representative Papers

Eco3S: Complex Socio-Economic System Simulation via Agent-Based Models

Jul 29, 2026

This work addresses the limitations of current large language model–driven agent-based modeling, which struggles to capture dynamically evolving agent–environment interactions, lacks counterfactual reasoning capabilities, and offers insufficient automation for scientific inquiry. To overcome these challenges, the authors propose a high-fidelity simulation framework tailored for socio-economic systems, innovatively integrating co-evolutionary mechanisms between agents and their environment, structural causal model (SCM)–based counterfactual reasoning, and a self-correcting paradigm that closes the loop among simulation, analysis, and optimization. By combining large language models with automated workflows, the framework enables end-to-end modeling. Empirical validation demonstrates its ability to successfully reproduce canonical economic phenomena—such as canal decline, the emergence of governance, and information diffusion—thereby confirming its intervenability, iterability, scalability, and generalizability across diverse scenarios.

0 citationsRead paper

Simple-to-Complex Structured Demonstrations for Vision-Language-Action Learning

Jul 05, 2026

This work addresses a critical yet overlooked aspect in vision-language-action (VLA) learning: the organization of demonstration data, which significantly impacts policy learning efficiency, stability, and generalization. The study pioneers the treatment of demonstration organization as a key design factor in VLA systems and introduces three general principles—task decomposition, environmental standardization, and curriculum-based data ordering by complexity—to construct a structured demonstration collection strategy. Implemented on a dual-arm robotic platform, this approach enables the model to progressively acquire fundamental skills from simple to complex scenarios and subsequently compose them to accomplish intricate tasks. Evaluated on block manipulation and towel folding benchmarks, the proposed method substantially outperforms end-to-end trajectory collection baselines in both task success rate and training stability.

0 citationsRead paper

Markovian Arrival Process Parameter Estimation of Quasi-birth-death Queueing Systems with Utilization Data

Jul 02, 2026

This work proposes a parameter estimation method based on the Expectation–Maximization (EM) algorithm for Markovian Arrival Process (MAP)-driven Quasi-Birth–Death (QBD) queueing systems, tailored to realistic scenarios where only coarse-grained data such as system utilization are available. Within a maximum likelihood framework, the approach infers sufficient statistics—including sojourn times, phase transitions, and service dynamics—underlying the hidden states directly from utilization time series. To the best of our knowledge, this is the first method capable of fully estimating MAP-QBD model parameters using solely utilization data. The study further introduces an innovative use of the Akaike Information Criterion (AIC) to automatically select the number of MAP phases, thereby mitigating overfitting. Experimental results demonstrate that the method accurately recovers both arrival and service parameters, offering a practical performance modeling tool for real-world systems lacking fine-grained event logs.

0 citationsRead paper

GenWorld: Empirically Grounded Urban Simulation Infrastructure for Scalable LLM-Agent Studies

Jun 25, 2026

This work addresses the limitations of large language model (LLM)-driven agents in urban-scale simulations, particularly concerning environmental fidelity and computational scalability. The authors propose a synthetic simulation framework grounded in real-world urban data, which uniquely integrates building-level urban modeling, empirically derived population distributions, and LLM-based agent behaviors. By offline-compiling LLM decisions into lookup-table policies, the framework preserves behavioral realism while enabling efficient large-scale simulation. The approach supports reproducible and auditable city-level experiments, demonstrated through a successful deployment in Higashihiroshima City, Hiroshima Prefecture, simulating 196,608 residents with validated demographic consistency and executing multi-scenario analyses—including typical weekdays, weekends, and emergency perturbations.

0 citationsRead paper

Polynomial Dice Loss for Medical Image Segmentation

Jun 22, 2026

This work addresses the performance bottlenecks in medical image segmentation caused by data imbalance and the difficulty of detecting small lesions. To this end, the authors propose a novel polynomial Dice loss function based on Taylor expansion. This approach reformulates the conventional Dice loss into a polynomial form with adjustable higher-order terms, enabling flexible control over each term’s contribution to the overall loss and thereby refining the geometric properties and optimization dynamics of the loss function. Experimental results across multiple medical image segmentation tasks demonstrate that the proposed method significantly outperforms standard Dice loss and its Tversky variant, achieving superior segmentation accuracy and robustness.

0 citationsRead paper

Recent publications

Latest Papers

Eco3S: Complex Socio-Economic System Simulation via Agent-Based Models

Jul 29, 2026

This work addresses the limitations of current large language model–driven agent-based modeling, which struggles to capture dynamically evolving agent–environment interactions, lacks counterfactual reasoning capabilities, and offers insufficient automation for scientific inquiry. To overcome these challenges, the authors propose a high-fidelity simulation framework tailored for socio-economic systems, innovatively integrating co-evolutionary mechanisms between agents and their environment, structural causal model (SCM)–based counterfactual reasoning, and a self-correcting paradigm that closes the loop among simulation, analysis, and optimization. By combining large language models with automated workflows, the framework enables end-to-end modeling. Empirical validation demonstrates its ability to successfully reproduce canonical economic phenomena—such as canal decline, the emergence of governance, and information diffusion—thereby confirming its intervenability, iterability, scalability, and generalizability across diverse scenarios.

0 citationsRead paper

Simple-to-Complex Structured Demonstrations for Vision-Language-Action Learning

Jul 05, 2026

This work addresses a critical yet overlooked aspect in vision-language-action (VLA) learning: the organization of demonstration data, which significantly impacts policy learning efficiency, stability, and generalization. The study pioneers the treatment of demonstration organization as a key design factor in VLA systems and introduces three general principles—task decomposition, environmental standardization, and curriculum-based data ordering by complexity—to construct a structured demonstration collection strategy. Implemented on a dual-arm robotic platform, this approach enables the model to progressively acquire fundamental skills from simple to complex scenarios and subsequently compose them to accomplish intricate tasks. Evaluated on block manipulation and towel folding benchmarks, the proposed method substantially outperforms end-to-end trajectory collection baselines in both task success rate and training stability.

0 citationsRead paper

Markovian Arrival Process Parameter Estimation of Quasi-birth-death Queueing Systems with Utilization Data

Jul 02, 2026

This work proposes a parameter estimation method based on the Expectation–Maximization (EM) algorithm for Markovian Arrival Process (MAP)-driven Quasi-Birth–Death (QBD) queueing systems, tailored to realistic scenarios where only coarse-grained data such as system utilization are available. Within a maximum likelihood framework, the approach infers sufficient statistics—including sojourn times, phase transitions, and service dynamics—underlying the hidden states directly from utilization time series. To the best of our knowledge, this is the first method capable of fully estimating MAP-QBD model parameters using solely utilization data. The study further introduces an innovative use of the Akaike Information Criterion (AIC) to automatically select the number of MAP phases, thereby mitigating overfitting. Experimental results demonstrate that the method accurately recovers both arrival and service parameters, offering a practical performance modeling tool for real-world systems lacking fine-grained event logs.

0 citationsRead paper

GenWorld: Empirically Grounded Urban Simulation Infrastructure for Scalable LLM-Agent Studies

Jun 25, 2026

This work addresses the limitations of large language model (LLM)-driven agents in urban-scale simulations, particularly concerning environmental fidelity and computational scalability. The authors propose a synthetic simulation framework grounded in real-world urban data, which uniquely integrates building-level urban modeling, empirically derived population distributions, and LLM-based agent behaviors. By offline-compiling LLM decisions into lookup-table policies, the framework preserves behavioral realism while enabling efficient large-scale simulation. The approach supports reproducible and auditable city-level experiments, demonstrated through a successful deployment in Higashihiroshima City, Hiroshima Prefecture, simulating 196,608 residents with validated demographic consistency and executing multi-scenario analyses—including typical weekdays, weekends, and emergency perturbations.

0 citationsRead paper

Polynomial Dice Loss for Medical Image Segmentation

Jun 22, 2026

This work addresses the performance bottlenecks in medical image segmentation caused by data imbalance and the difficulty of detecting small lesions. To this end, the authors propose a novel polynomial Dice loss function based on Taylor expansion. This approach reformulates the conventional Dice loss into a polynomial form with adjustable higher-order terms, enabling flexible control over each term’s contribution to the overall loss and thereby refining the geometric properties and optimization dynamics of the loss function. Experimental results across multiple medical image segmentation tasks demonstrate that the proposed method significantly outperforms standard Dice loss and its Tversky variant, achieving superior segmentation accuracy and robustness.

0 citationsRead paper