Selected work

Representative Papers

No Data Is Not No Risk: Visibility Aware Graph-Based Inference of Business Conduct Risk

Jul 29, 2026

This study addresses the challenge of corporate risk monitoring, which is often hindered by data sparsity, imbalanced distributions, and visibility bias—where unrecorded events do not necessarily imply absence of risk. The problem is formulated as a positive-unlabeled (PU) node classification task on an equity ownership graph. To tackle potential contamination from latent positive samples in the unlabeled set, the authors propose a GCNII-based framework that integrates relation-aware message passing with non-negative PU learning. By constructing a corporate ownership graph and incorporating relation-specific message propagation mechanisms, the model achieves significantly superior performance in prospective risk assessment compared to both non-graph and baseline graph-based methods, demonstrating robust predictive capability even for firms with no prior risk records.

0 citationsRead paper

Interdomain Attention: Beyond Token-Level Key-Value Memory

May 22, 2026

This work addresses the high computational cost of standard Transformer attention and the limited query-conditioned expressivity of state space models (SSMs), which, despite their length-independent scalability, lack content-aware matching. The authors propose Interdomain Attention, a novel architecture that embeds SSMs into attention modules via kernel methods: by approximating the attention kernel with finite feature maps, keys and values are projected onto a shared set of SSM-maintained basis functions, while queries attend to the compressed coefficients through their own feature mappings. This enables query-conditioned attention within a fixed-size state for the first time. Combining the content-matching strength of attention with the scalability of SSMs, the method consistently outperforms pure SSM mixers under identical state budgets across language models ranging from 125M to 1.3B parameters. The 1.3B variant surpasses standard softmax attention baselines in validation perplexity and eight commonsense reasoning tasks, maintaining stable performance even when trained on contexts 3.5× longer.

0 citationsRead paper

Multiscale Euclidean Network Trajectories: Second-Moment Geometry, Attribution, and Change Points

May 06, 2026

A central challenge in dynamic network analysis is to represent temporal evolution in a way that is both geometrically meaningful and statistically identifiable. One approach embeds a sequence of network snapshots as trajectories in a Euclidean space and relates these trajectories to node embeddings. In multilayer and unfolded spectral constructions, however, node embeddings and their underlying latent positions are identifiable only up to general linear transformations. Although this ambiguity preserves edge probabilities, it can distort geometry and invalidate distance based temporal comparisons at both the trajectory and node-levels. We develop Multiscale Euclidean Network Trajectories (MENT), a framework for multiscale temporal trajectories based on second-moment geometry. By imposing an isotropic normalization on the anchor latent positions, we reduce the relevant ambiguity to orthogonal transformations and prevent distortion of the second-moment geometry. In this canonical representation, we define a trace variation distance and mode-wise variation distances along orthogonal directions, and use multidimensional scaling to obtain low-dimensional trajectories of time points at both global and mode-wise levels. The resulting trajectories support interpretation and inference. They admit mode-wise decompositions, support attribution of global and mode-wise temporal changes to nodes, and enable change point detection through 1D trajectories. We prove consistency of the proposed unfolded spectral embedding and of the induced temporal trajectories. Experiments on two synthetic and two real dynamic networks illustrate stable and interpretable recovery of temporal structure and show strong performance against existing change point detection baselines.

0 citationsRead paper

Deep Learning-Based Airway Segmentation in Systemic Lupus Erythematosus Patients with Interstitial Lung Disease (SLE-ILD): A Comparative High-Resolution CT Analysis

Mar 18, 2026

This study addresses the identification of airway structural alterations in patients with systemic lupus erythematosus-associated interstitial lung disease (SLE-ILD) at the lobar and segmental levels. Leveraging high-resolution computed tomography (HRCT) images, we developed a customized U-Net deep learning model to enable automated segmentation and quantitative volumetric analysis of airways within individual lung lobes and segments. For the first time using an AI-driven approach, we identified a regional airway dilation phenotype predominantly affecting the upper lung zones—specifically the right and left upper lobes and the R1, R3, and L3 bronchopulmonary segments—in SLE-ILD patients (p<0.05). These findings suggest that airway volume may serve as a potential imaging biomarker, offering a novel avenue for the early detection and monitoring of SLE-ILD.

0 citationsRead paper

Magnetic Resonance Simulation of Effective Transverse Relaxation (T2*)

Jan 27, 2026

This work proposes an efficient method for simulating the reversible component of T2*, denoted T2', in conventional magnetic resonance simulations. Traditionally, accurately approximating the Lorentzian line shape of T2' requires a large number (>100) of isochromats, resulting in high computational cost. The proposed approach leverages a linear phase model to directly characterize the Lorentzian response of T2' by simultaneously simulating the frequency derivative of magnetization and integrating analytical solutions with joint transition techniques to accelerate computation. Remarkably, this method achieves accurate T2' simulation using only a single isochromat, enabling high-fidelity reconstruction in standard pulse sequences with only a 2.0–2.7× increase in overall computational overhead. The analytical solution and joint transition strategy contribute speedups of up to 19× and 17×, respectively. This study represents the first integration of the linear phase model with derivative-based simulation for T2' modeling, substantially improving efficiency without compromising accuracy.

0 citationsRead paper

Recent publications

Latest Papers

No Data Is Not No Risk: Visibility Aware Graph-Based Inference of Business Conduct Risk

Jul 29, 2026

This study addresses the challenge of corporate risk monitoring, which is often hindered by data sparsity, imbalanced distributions, and visibility bias—where unrecorded events do not necessarily imply absence of risk. The problem is formulated as a positive-unlabeled (PU) node classification task on an equity ownership graph. To tackle potential contamination from latent positive samples in the unlabeled set, the authors propose a GCNII-based framework that integrates relation-aware message passing with non-negative PU learning. By constructing a corporate ownership graph and incorporating relation-specific message propagation mechanisms, the model achieves significantly superior performance in prospective risk assessment compared to both non-graph and baseline graph-based methods, demonstrating robust predictive capability even for firms with no prior risk records.

0 citationsRead paper

Interdomain Attention: Beyond Token-Level Key-Value Memory

May 22, 2026

This work addresses the high computational cost of standard Transformer attention and the limited query-conditioned expressivity of state space models (SSMs), which, despite their length-independent scalability, lack content-aware matching. The authors propose Interdomain Attention, a novel architecture that embeds SSMs into attention modules via kernel methods: by approximating the attention kernel with finite feature maps, keys and values are projected onto a shared set of SSM-maintained basis functions, while queries attend to the compressed coefficients through their own feature mappings. This enables query-conditioned attention within a fixed-size state for the first time. Combining the content-matching strength of attention with the scalability of SSMs, the method consistently outperforms pure SSM mixers under identical state budgets across language models ranging from 125M to 1.3B parameters. The 1.3B variant surpasses standard softmax attention baselines in validation perplexity and eight commonsense reasoning tasks, maintaining stable performance even when trained on contexts 3.5× longer.

0 citationsRead paper

Multiscale Euclidean Network Trajectories: Second-Moment Geometry, Attribution, and Change Points

May 06, 2026

A central challenge in dynamic network analysis is to represent temporal evolution in a way that is both geometrically meaningful and statistically identifiable. One approach embeds a sequence of network snapshots as trajectories in a Euclidean space and relates these trajectories to node embeddings. In multilayer and unfolded spectral constructions, however, node embeddings and their underlying latent positions are identifiable only up to general linear transformations. Although this ambiguity preserves edge probabilities, it can distort geometry and invalidate distance based temporal comparisons at both the trajectory and node-levels. We develop Multiscale Euclidean Network Trajectories (MENT), a framework for multiscale temporal trajectories based on second-moment geometry. By imposing an isotropic normalization on the anchor latent positions, we reduce the relevant ambiguity to orthogonal transformations and prevent distortion of the second-moment geometry. In this canonical representation, we define a trace variation distance and mode-wise variation distances along orthogonal directions, and use multidimensional scaling to obtain low-dimensional trajectories of time points at both global and mode-wise levels. The resulting trajectories support interpretation and inference. They admit mode-wise decompositions, support attribution of global and mode-wise temporal changes to nodes, and enable change point detection through 1D trajectories. We prove consistency of the proposed unfolded spectral embedding and of the induced temporal trajectories. Experiments on two synthetic and two real dynamic networks illustrate stable and interpretable recovery of temporal structure and show strong performance against existing change point detection baselines.

0 citationsRead paper

Deep Learning-Based Airway Segmentation in Systemic Lupus Erythematosus Patients with Interstitial Lung Disease (SLE-ILD): A Comparative High-Resolution CT Analysis

Mar 18, 2026

This study addresses the identification of airway structural alterations in patients with systemic lupus erythematosus-associated interstitial lung disease (SLE-ILD) at the lobar and segmental levels. Leveraging high-resolution computed tomography (HRCT) images, we developed a customized U-Net deep learning model to enable automated segmentation and quantitative volumetric analysis of airways within individual lung lobes and segments. For the first time using an AI-driven approach, we identified a regional airway dilation phenotype predominantly affecting the upper lung zones—specifically the right and left upper lobes and the R1, R3, and L3 bronchopulmonary segments—in SLE-ILD patients (p<0.05). These findings suggest that airway volume may serve as a potential imaging biomarker, offering a novel avenue for the early detection and monitoring of SLE-ILD.

0 citationsRead paper

Magnetic Resonance Simulation of Effective Transverse Relaxation (T2*)

Jan 27, 2026

This work proposes an efficient method for simulating the reversible component of T2*, denoted T2', in conventional magnetic resonance simulations. Traditionally, accurately approximating the Lorentzian line shape of T2' requires a large number (>100) of isochromats, resulting in high computational cost. The proposed approach leverages a linear phase model to directly characterize the Lorentzian response of T2' by simultaneously simulating the frequency derivative of magnetization and integrating analytical solutions with joint transition techniques to accelerate computation. Remarkably, this method achieves accurate T2' simulation using only a single isochromat, enabling high-fidelity reconstruction in standard pulse sequences with only a 2.0–2.7× increase in overall computational overhead. The analytical solution and joint transition strategy contribute speedups of up to 19× and 17×, respectively. This study represents the first integration of the linear phase model with derivative-based simulation for T2' modeling, substantially improving efficiency without compromising accuracy.

0 citationsRead paper