Self-Consistent Adjoint Policy Iteration for Constrained Dynamic Portfolio Choice
研究通过自洽伴随策略迭代方法解决了带约束的动态投资组合选择问题，适用于可预测回报和凸约束情况。
0 citationsRead paper
Selected work
Representative Papers
Scalable Pontryagin-Guided Adjoint-to-Control Recovery for Constrained Dynamic Portfolio Choice
This study addresses the challenges of adjoint variable recovery and scalable control in continuous-time dynamic portfolio selection with smoothing constraints by proposing an "Adjoint-to-Control" framework. By establishing a theoretical correspondence between Open-Loop Backpropagation Through Time (OL-BPTT) and Pontryagin’s Maximum Principle (PMP), and integrating Differentiable Path Optimization (DPO) for path sensitivity, the method employs nested regression to estimate martingale inputs and solve local generalized Hamiltonian problems. This approach enables efficient and precise recovery of optimal strategies under high-dimensional constraints. Benchmarks involving one hundred assets demonstrate an adjoint error of only 0.46% and a strategy RMSE below 8.5×10⁻³, while significantly reducing KKT residuals. These results validate the method's accuracy and practicality for high-dimensional financial control applications.
0 citationsRead paper
Characterising cardiac tissue properties with graph neural networks
This study addresses the challenges of cardiac electrophysiological characterization and ablation target localization under sparse intracardiac measurements by proposing a graph neural network-based framework for sparse data representation. Through pre-training on synthetic signals to identify regions of interest for premature ventricular contraction ablation, combined with few-shot fine-tuning, the method achieves cross-domain generalization from planar to curved surfaces. Experimental results demonstrate average precisions of 0.96, 0.97, and 0.95 in detecting fibrosis, rapid depolarization, and high excitability, respectively. These findings indicate that the proposed approach effectively enables precise mapping despite clinical data sparsity, exhibiting superior generalization capabilities and significant potential for clinical translation in cardiac electrophysiology procedures.
0 citationsRead paper
Fully Automated End-to-End Adversary Emulation from MITRE ATT\&CK Based Cyber Threat Intelligence Using LLMs
This work proposes an end-to-end automated red-teaming framework that overcomes the limitations of existing adversary emulation approaches, which typically rely on predefined playbooks or manual intervention and struggle to automatically derive executable attack procedures from cyber threat intelligence (CTI). The proposed system uniquely integrates MITRE ATT&CK-aligned CTI report parsing, large language model–driven attack playbook generation, and an automatic execution pipeline with failure-type-aware repair mechanisms—all within a unified workflow requiring no human involvement. Implemented on the CALDERA platform and leveraging models such as Claude Sonnet 4.5, the framework generates an average of 27.3 attack capabilities per CTI report across 11 reports, achieving an 84.22% post-repair execution success rate and an F1 score of 60.50%, significantly outperforming AURORA.
0 citationsRead paper
The Circumplex Degeneracy Behind the Rare-Class Limit in Affect Recognition
This study addresses the poor performance of in-the-wild facial expression recognition on rare emotion categories, arguing that the issue stems not merely from data imbalance but from representational degeneration within the circular affective space defined by Russell’s circumplex model. The authors propose a novel approach grounded in optimal transport theory to construct an emotion-distance-based cost function, integrated with action unit augmentation and uniform cost regularization. Multi-task ablation experiments on AffectNet and Aff-Wild2 reveal that the core bottleneck lies in the geometric degradation of the affective space. The work further introduces a new evaluation paradigm distinguishing representational capacity from confusion-aware reweighting. Results show that while the proposed circular cost improves standard metrics, uniform costing yields superior performance on AffectNet; predictions on Aff-Wild2 better reflect true affective distributions, yet fine-grained distinctions—such as between anger and fear—remain challenging.
0 citationsRead paper
Recent publications
Latest Papers
Self-Consistent Adjoint Policy Iteration for Constrained Dynamic Portfolio Choice
研究通过自洽伴随策略迭代方法解决了带约束的动态投资组合选择问题，适用于可预测回报和凸约束情况。
0 citationsRead paper
Scalable Pontryagin-Guided Adjoint-to-Control Recovery for Constrained Dynamic Portfolio Choice
This study addresses the challenges of adjoint variable recovery and scalable control in continuous-time dynamic portfolio selection with smoothing constraints by proposing an "Adjoint-to-Control" framework. By establishing a theoretical correspondence between Open-Loop Backpropagation Through Time (OL-BPTT) and Pontryagin’s Maximum Principle (PMP), and integrating Differentiable Path Optimization (DPO) for path sensitivity, the method employs nested regression to estimate martingale inputs and solve local generalized Hamiltonian problems. This approach enables efficient and precise recovery of optimal strategies under high-dimensional constraints. Benchmarks involving one hundred assets demonstrate an adjoint error of only 0.46% and a strategy RMSE below 8.5×10⁻³, while significantly reducing KKT residuals. These results validate the method's accuracy and practicality for high-dimensional financial control applications.
0 citationsRead paper
Characterising cardiac tissue properties with graph neural networks
This study addresses the challenges of cardiac electrophysiological characterization and ablation target localization under sparse intracardiac measurements by proposing a graph neural network-based framework for sparse data representation. Through pre-training on synthetic signals to identify regions of interest for premature ventricular contraction ablation, combined with few-shot fine-tuning, the method achieves cross-domain generalization from planar to curved surfaces. Experimental results demonstrate average precisions of 0.96, 0.97, and 0.95 in detecting fibrosis, rapid depolarization, and high excitability, respectively. These findings indicate that the proposed approach effectively enables precise mapping despite clinical data sparsity, exhibiting superior generalization capabilities and significant potential for clinical translation in cardiac electrophysiology procedures.
0 citationsRead paper
Fully Automated End-to-End Adversary Emulation from MITRE ATT\&CK Based Cyber Threat Intelligence Using LLMs
This work proposes an end-to-end automated red-teaming framework that overcomes the limitations of existing adversary emulation approaches, which typically rely on predefined playbooks or manual intervention and struggle to automatically derive executable attack procedures from cyber threat intelligence (CTI). The proposed system uniquely integrates MITRE ATT&CK-aligned CTI report parsing, large language model–driven attack playbook generation, and an automatic execution pipeline with failure-type-aware repair mechanisms—all within a unified workflow requiring no human involvement. Implemented on the CALDERA platform and leveraging models such as Claude Sonnet 4.5, the framework generates an average of 27.3 attack capabilities per CTI report across 11 reports, achieving an 84.22% post-repair execution success rate and an F1 score of 60.50%, significantly outperforming AURORA.
0 citationsRead paper
The Circumplex Degeneracy Behind the Rare-Class Limit in Affect Recognition
This study addresses the poor performance of in-the-wild facial expression recognition on rare emotion categories, arguing that the issue stems not merely from data imbalance but from representational degeneration within the circular affective space defined by Russell’s circumplex model. The authors propose a novel approach grounded in optimal transport theory to construct an emotion-distance-based cost function, integrated with action unit augmentation and uniform cost regularization. Multi-task ablation experiments on AffectNet and Aff-Wild2 reveal that the core bottleneck lies in the geometric degradation of the affective space. The work further introduces a new evaluation paradigm distinguishing representational capacity from confusion-aware reweighting. Results show that while the proposed circular cost improves standard metrics, uniform costing yields superior performance on AffectNet; predictions on Aff-Wild2 better reflect true affective distributions, yet fine-grained distinctions—such as between anger and fear—remain challenging.
0 citationsRead paper