Selected work

Representative Papers

Conditional Neural Optimal Transport for Predicting Cellular Phenotypes from Molecular Structure

Aug 14, 2026

This study addresses the challenge of cellular phenotypic characterization arising from the vastness of chemical space by proposing a Molecule-Conditioned Neural Optimal Transport model. Conditioning on molecular structures, this approach integrates Monge-gap regularization with compressed representation space transport to learn mappings from negative controls to perturbed phenotypes, effectively overcoming static coupling failures in large-scale datasets. Experimental results demonstrate that the proposed framework significantly outperforms baseline methods on unseen active molecules by accurately recovering specific phenotypes and mitigating technical variation. Furthermore, it enables cross-batch phenotypic comparison and zero-shot generalization, validating its efficacy for virtual screening applications in drug discovery.

0 citationsRead paper

A Two-Regime Statistical Framework for Wind-Power Distributions: From Wind-Speed Fluctuations to Turbine Control

Jul 28, 2026

Wind power exhibits significant fluctuations due to the nonlinear relationship between wind speed and turbine output, posing challenges for reliable grid integration. This work proposes a dual-mechanism statistical framework that explicitly distinguishes two physically distinct operational regimes. In the aerodynamic region, the power distribution is analytically derived using a Rician wind speed model combined with the cubic wind speed–power relationship. In the near-rated control region, a bounded stretched exponential distribution is employed to capture the continuous tail behavior of power deficits. For the first time, turbine operation is partitioned into these two mechanistically interpretable intervals, each modeled with physically grounded statistical formulations to jointly characterize the full-range power distribution. Validation against empirical data demonstrates that the approach delivers high-accuracy, interpretable modeling of power variability for both individual turbines and wind farm clusters, substantially enhancing wind power forecasting and grid dispatch reliability.

0 citationsRead paper

Stakeholder Suite: A Unified AI Framework for Mapping Actors, Topics and Arguments in Public Debates

Dec 19, 2025

Public debates surrounding energy and infrastructure projects exhibit complex, evolving controversy dynamics; however, existing media analysis tools lack interpretability and fine-grained modeling capabilities. To address this, we propose the first multi-task collaborative AI framework tailored for public debate analysis, integrating named entity recognition, dynamic topic modeling, structured argument extraction, and multi-granularity stance classification. Our approach combines fine-tuned language models with rule-based enhancements to construct source-text-anchored debate graphs. The framework demonstrates strong domain generalizability and practical deployability: across multiple energy projects, it achieves a 75% human-validated argument relevance rate. It significantly improves timeliness in controversy anticipation, accuracy in stance identification, and clarity in influence network visualization—capabilities already operationalized in real-world decision-support systems.

0 citationsRead paper

A note on a resampling procedure for estimating the density at a given quantile

Sep 02, 2025

This paper addresses the problem of density estimation at a specified quantile. We propose a novel resampling-based method: multiple zero-mean Gaussian variates are generated, and a least-squares estimator is directly constructed at the target quantile, achieving parametric convergence rates. Theoretical analysis reveals the critical role of the Gaussian sampling variance in estimation accuracy and establishes sufficient conditions for estimator consistency. Furthermore, an adaptive grid search algorithm is designed to automatically select the optimal variance. Compared with conventional kernel density estimation, our method demonstrates significantly improved estimation accuracy and faster convergence in simulation studies, while retaining rigorous theoretical guarantees and computational feasibility.

0 citationsRead paper

Dragonfly: a modular deep reinforcement learning library

Apr 30, 2025

To address low algorithm reusability, high configuration maintenance overhead, and poor runtime efficiency of CPU-intensive simulation environments in deep reinforcement learning (DRL) experiments, this paper introduces a lightweight, modular open-source DRL library. Methodologically, it proposes a configuration-driven architecture based on JSON serialization, enabling full decoupling and zero-code switching of algorithmic components—including neural network architectures, samplers, and optimizers—and designs an efficient execution engine optimized for CPU-bound simulations via intelligent resource scheduling and batched inference. The library ships with standardized implementations of DQN and PPO, along with seamless integrations for Atari and MuJoCo benchmarks. Empirical evaluation demonstrates competitive or superior performance against RLlib and Stable-Baselines3 on standard benchmarks, while reducing configuration modification time by 90% and increasing simulation task throughput by 40%.

0 citationsRead paper

Recent publications

Latest Papers

Conditional Neural Optimal Transport for Predicting Cellular Phenotypes from Molecular Structure

Aug 14, 2026

This study addresses the challenge of cellular phenotypic characterization arising from the vastness of chemical space by proposing a Molecule-Conditioned Neural Optimal Transport model. Conditioning on molecular structures, this approach integrates Monge-gap regularization with compressed representation space transport to learn mappings from negative controls to perturbed phenotypes, effectively overcoming static coupling failures in large-scale datasets. Experimental results demonstrate that the proposed framework significantly outperforms baseline methods on unseen active molecules by accurately recovering specific phenotypes and mitigating technical variation. Furthermore, it enables cross-batch phenotypic comparison and zero-shot generalization, validating its efficacy for virtual screening applications in drug discovery.

0 citationsRead paper

A Two-Regime Statistical Framework for Wind-Power Distributions: From Wind-Speed Fluctuations to Turbine Control

Jul 28, 2026

Wind power exhibits significant fluctuations due to the nonlinear relationship between wind speed and turbine output, posing challenges for reliable grid integration. This work proposes a dual-mechanism statistical framework that explicitly distinguishes two physically distinct operational regimes. In the aerodynamic region, the power distribution is analytically derived using a Rician wind speed model combined with the cubic wind speed–power relationship. In the near-rated control region, a bounded stretched exponential distribution is employed to capture the continuous tail behavior of power deficits. For the first time, turbine operation is partitioned into these two mechanistically interpretable intervals, each modeled with physically grounded statistical formulations to jointly characterize the full-range power distribution. Validation against empirical data demonstrates that the approach delivers high-accuracy, interpretable modeling of power variability for both individual turbines and wind farm clusters, substantially enhancing wind power forecasting and grid dispatch reliability.

0 citationsRead paper

Stakeholder Suite: A Unified AI Framework for Mapping Actors, Topics and Arguments in Public Debates

Dec 19, 2025

Public debates surrounding energy and infrastructure projects exhibit complex, evolving controversy dynamics; however, existing media analysis tools lack interpretability and fine-grained modeling capabilities. To address this, we propose the first multi-task collaborative AI framework tailored for public debate analysis, integrating named entity recognition, dynamic topic modeling, structured argument extraction, and multi-granularity stance classification. Our approach combines fine-tuned language models with rule-based enhancements to construct source-text-anchored debate graphs. The framework demonstrates strong domain generalizability and practical deployability: across multiple energy projects, it achieves a 75% human-validated argument relevance rate. It significantly improves timeliness in controversy anticipation, accuracy in stance identification, and clarity in influence network visualization—capabilities already operationalized in real-world decision-support systems.

0 citationsRead paper

A note on a resampling procedure for estimating the density at a given quantile

Sep 02, 2025

This paper addresses the problem of density estimation at a specified quantile. We propose a novel resampling-based method: multiple zero-mean Gaussian variates are generated, and a least-squares estimator is directly constructed at the target quantile, achieving parametric convergence rates. Theoretical analysis reveals the critical role of the Gaussian sampling variance in estimation accuracy and establishes sufficient conditions for estimator consistency. Furthermore, an adaptive grid search algorithm is designed to automatically select the optimal variance. Compared with conventional kernel density estimation, our method demonstrates significantly improved estimation accuracy and faster convergence in simulation studies, while retaining rigorous theoretical guarantees and computational feasibility.

0 citationsRead paper

Dragonfly: a modular deep reinforcement learning library

Apr 30, 2025

To address low algorithm reusability, high configuration maintenance overhead, and poor runtime efficiency of CPU-intensive simulation environments in deep reinforcement learning (DRL) experiments, this paper introduces a lightweight, modular open-source DRL library. Methodologically, it proposes a configuration-driven architecture based on JSON serialization, enabling full decoupling and zero-code switching of algorithmic components—including neural network architectures, samplers, and optimizers—and designs an efficient execution engine optimized for CPU-bound simulations via intelligent resource scheduling and batched inference. The library ships with standardized implementations of DQN and PPO, along with seamless integrations for Atari and MuJoCo benchmarks. Empirical evaluation demonstrates competitive or superior performance against RLlib and Stable-Baselines3 on standard benchmarks, while reducing configuration modification time by 90% and increasing simulation task throughput by 40%.

0 citationsRead paper