Scalability Analysis of Distributed Kolmogorov-Arnold Network Training on High-Performance Computing Systems
研究通过在高性能计算系统上使用分布式Kolmogorov-Arnold网络训练，探讨了其可扩展性问题，采用多节点多GPU设置进行实验分析。
0 citationsRead paper
Selected work
Representative Papers
Uncertainty-Aware Missing-Data Multimodal Latent for Fetal-Growth Analysis
This study addresses the challenge of missing multimodal measurements and low inter-modality redundancy in clinical fetal growth assessment by proposing a linear Gaussian factor model that leverages marginalization rather than imputation. The model jointly represents four classes of fetal and maternal data, naturally encoding observation completeness and yielding uncertainty-aware latent representations. Factor dimensionality is determined via parallel analysis, with K=8 factors subjected to VARIMAX rotation, and anomalous records are flagged using standardized residuals. Empirical results demonstrate near-independence across modalities (cross-modal R² = 0.023), a significant correlation between latent dimensions and birth weight percentile (ρ = 0.55), and an AUC of 0.70 for predicting small-for-gestational-age outcomes along a hemodynamic axis. The model achieves nominal 97% confidence interval coverage and successfully identifies 36 data entry errors.
0 citationsRead paper
Toward cryptographically verifiable authorization for autonomous AI agents: A security hypothesis, preliminary formal model, and proof-of-concept implementation
This work addresses the absence of cryptographic evidence verifying that autonomous AI agents comply with authorization policies within specific execution contexts. To bridge this gap, the paper introduces a cryptographically verifiable authorization relation, denoted $R_{CVA}$, which structurally decouples identity, request, and execution context. It formally defines critical security properties—such as authorization soundness and policy binding—and enables verification while preserving the confidentiality of private attributes. The authors instantiate the core model through a zero-knowledge proof prototype built upon the Groth16 zk-SNARK framework, thereby providing the first practical demonstration of cryptographically verifiable agent authorization.
0 citationsRead paper
Sensing the properties of virtual objects without physical feedback
In extended reality (XR) environments lacking physical haptic feedback, how users perceive the stiffness of virtual objects—such as C$_{60}$ molecules—remains unclear. This study addresses this gap by integrating an interactive molecular dynamics XR system (iMD-XR), psychophysical experiments, and systematic stiffness parameter modulation to quantify, for the first time, user perceptual thresholds for virtual molecular stiffness in the absence of tactile cues. Results demonstrate that the just-noticeable difference (JND) under direct interaction is 11.5%, significantly lower than the 18.5% observed under passive viewing alone. Moreover, engaging in interaction prior to observation markedly enhances perceptual accuracy during subsequent viewing. These findings reveal the facilitative role of active interaction in perceiving virtual object properties and demonstrate a cross-modal transfer effect from action to perception.
0 citationsRead paper
A new classification method based on Minimum Spanning Trees
This work presents the first systematic integration of minimum spanning trees (MSTs) into supervised classification tasks, addressing their applicability and computational efficiency in such settings. To overcome the limitations of conventional MST-based approaches in the presence of noise and high-dimensional data, the authors propose a robust and computationally efficient MST classification algorithm. By synergistically combining graph theory with supervised learning and incorporating structural optimizations, the method enhances generalization performance. Extensive experiments on large-scale synthetic datasets and real-world aviation trajectory data demonstrate that the proposed algorithm achieves high classification accuracy while significantly outperforming baseline methods, thereby confirming its effectiveness and practical utility.
0 citationsRead paper
Recent publications
Latest Papers
Scalability Analysis of Distributed Kolmogorov-Arnold Network Training on High-Performance Computing Systems
研究通过在高性能计算系统上使用分布式Kolmogorov-Arnold网络训练，探讨了其可扩展性问题，采用多节点多GPU设置进行实验分析。
0 citationsRead paper
Uncertainty-Aware Missing-Data Multimodal Latent for Fetal-Growth Analysis
This study addresses the challenge of missing multimodal measurements and low inter-modality redundancy in clinical fetal growth assessment by proposing a linear Gaussian factor model that leverages marginalization rather than imputation. The model jointly represents four classes of fetal and maternal data, naturally encoding observation completeness and yielding uncertainty-aware latent representations. Factor dimensionality is determined via parallel analysis, with K=8 factors subjected to VARIMAX rotation, and anomalous records are flagged using standardized residuals. Empirical results demonstrate near-independence across modalities (cross-modal R² = 0.023), a significant correlation between latent dimensions and birth weight percentile (ρ = 0.55), and an AUC of 0.70 for predicting small-for-gestational-age outcomes along a hemodynamic axis. The model achieves nominal 97% confidence interval coverage and successfully identifies 36 data entry errors.
0 citationsRead paper
Toward cryptographically verifiable authorization for autonomous AI agents: A security hypothesis, preliminary formal model, and proof-of-concept implementation
This work addresses the absence of cryptographic evidence verifying that autonomous AI agents comply with authorization policies within specific execution contexts. To bridge this gap, the paper introduces a cryptographically verifiable authorization relation, denoted $R_{CVA}$, which structurally decouples identity, request, and execution context. It formally defines critical security properties—such as authorization soundness and policy binding—and enables verification while preserving the confidentiality of private attributes. The authors instantiate the core model through a zero-knowledge proof prototype built upon the Groth16 zk-SNARK framework, thereby providing the first practical demonstration of cryptographically verifiable agent authorization.
0 citationsRead paper
Sensing the properties of virtual objects without physical feedback
In extended reality (XR) environments lacking physical haptic feedback, how users perceive the stiffness of virtual objects—such as C$_{60}$ molecules—remains unclear. This study addresses this gap by integrating an interactive molecular dynamics XR system (iMD-XR), psychophysical experiments, and systematic stiffness parameter modulation to quantify, for the first time, user perceptual thresholds for virtual molecular stiffness in the absence of tactile cues. Results demonstrate that the just-noticeable difference (JND) under direct interaction is 11.5%, significantly lower than the 18.5% observed under passive viewing alone. Moreover, engaging in interaction prior to observation markedly enhances perceptual accuracy during subsequent viewing. These findings reveal the facilitative role of active interaction in perceiving virtual object properties and demonstrate a cross-modal transfer effect from action to perception.
0 citationsRead paper
A new classification method based on Minimum Spanning Trees
This work presents the first systematic integration of minimum spanning trees (MSTs) into supervised classification tasks, addressing their applicability and computational efficiency in such settings. To overcome the limitations of conventional MST-based approaches in the presence of noise and high-dimensional data, the authors propose a robust and computationally efficient MST classification algorithm. By synergistically combining graph theory with supervised learning and incorporating structural optimizations, the method enhances generalization performance. Extensive experiments on large-scale synthetic datasets and real-world aviation trajectory data demonstrate that the proposed algorithm achieves high classification accuracy while significantly outperforming baseline methods, thereby confirming its effectiveness and practical utility.
0 citationsRead paper