Building Trustworthy Graph-Agentic RAG for Social Good: Architectures, Failure Propagation, and Assurance by Construction
论文探讨了如何设计和评估图-代理检索增强生成系统，以确保其在社会公益场景中的可信度，通过构建五个接口契约来解决耦合失败路径问题。
0 citationsRead paper
Selected work
Representative Papers
Short Cycles Decide P-versus-NPC Status ofHamiltonicity on Bisplit Graphs
This study delineates the computational complexity boundary of the Hamiltonian cycle and path problems on split–split graphs. By introducing chordality and forbidden induced path length (Pₖ-free) as key structural parameters, and leveraging graph decomposition, structural graph theory, and complexity reductions, the work precisely identifies the threshold at which the problem transitions from polynomial-time solvability to NP-completeness. The main contributions include proving that the problem is polynomial-time solvable on chordal split–split graphs, yet NP-complete even on chordal bipartite split–split graphs. Furthermore, the paper establishes a tight complexity dichotomy by showing tractability for P₅-free instances and intractability for P₁₀-free instances, extending these results to several related variants.
0 citationsRead paper
Multi-Modal Machine Learning for Population- and Subject-Specific lncRNA-Type 2 Diabetes Association Analysis
Traditional single-omics approaches struggle to fully elucidate the complex regulatory mechanisms of long non-coding RNAs (lncRNAs) in type 2 diabetes (T2D). This study proposes the first multimodal machine learning framework integrating lncRNA expression profiles, secondary structures, and sequence features. By combining eight classifiers, hierarchical cross-validation, and SHAP-based interpretability analysis, the framework enables precise association mapping from population to individual levels across two independent cohorts. The approach identifies multiple T2D-significant lncRNAs—including GAS5, XIST, MEG3, and ANRIL—with MEG3 consistently emerging as the top cross-cohort driver according to SHAP values. These findings not only corroborate results from conventional statistical methods but also yield a higher-resolution regulatory landscape, advancing the potential of lncRNAs in T2D precision medicine.
0 citationsRead paper
PhaseNet++: Phase-Aware Frequency-Domain Anomaly Detection for Industrial Control Systems via Phase Coherence Graphs
This work addresses the common oversight of phase information in the frequency domain for multivariate time series anomaly detection in industrial control systems, which often hinders the identification of synchronized anomalies across sensors. To this end, the authors propose PhaseNet++, a novel framework that systematically incorporates phase cues by extracting magnitude and phase spectra via short-time Fourier transform, constructing a lightweight graph structure using phase coherence indices, and jointly reconstructing signals through an integrated architecture combining graph attention networks and a Sensor-Token Transformer. A dual-head decoder—featuring recurrent and coherence-aware branches—is further introduced to enhance modeling capacity. Evaluated on the SWaT dataset, PhaseNet++ achieves an F1-score of 90.98%, ROC-AUC of 95.66%, and average precision of 91.51%, substantially outperforming existing methods with only 264,816 additional parameters.
0 citationsRead paper
ioPUF+: A PUF Based on I/O Pull-Up/Down Resistors for Secret Key Generation in IoT Nodes
To address the demand for low-cost, off-the-shelf hardware–compatible Physical Unclonable Functions (PUFs) in IoT node key generation, this paper proposes a lightweight PUF mechanism leveraging process variations in commercial chips’ I/O pull-up/pull-down resistors. The approach requires no IC design or fabrication modifications and extracts device-unique fingerprints solely from standard I/O structures. It enables end-to-end conversion from raw responses to cryptographically secure keys via resistance measurement, BCH error correction, and SHA-256 hashing, integrated within an AES-secured communication pipeline. Evaluated on 30 commercial MCUs, the PUF achieves 100% intra-chip Hamming distance consistency, 50.33% inter-chip uniqueness, 50.54% response uniformity, and a worst-case bit error rate <2.63%. The implementation occupies only 19.8 KB of flash memory, consumes 79 mW, and incurs a key generation latency of 600 ms.
0 citationsRead paper
Recent publications
Latest Papers
Building Trustworthy Graph-Agentic RAG for Social Good: Architectures, Failure Propagation, and Assurance by Construction
论文探讨了如何设计和评估图-代理检索增强生成系统，以确保其在社会公益场景中的可信度，通过构建五个接口契约来解决耦合失败路径问题。
0 citationsRead paper
Short Cycles Decide P-versus-NPC Status ofHamiltonicity on Bisplit Graphs
This study delineates the computational complexity boundary of the Hamiltonian cycle and path problems on split–split graphs. By introducing chordality and forbidden induced path length (Pₖ-free) as key structural parameters, and leveraging graph decomposition, structural graph theory, and complexity reductions, the work precisely identifies the threshold at which the problem transitions from polynomial-time solvability to NP-completeness. The main contributions include proving that the problem is polynomial-time solvable on chordal split–split graphs, yet NP-complete even on chordal bipartite split–split graphs. Furthermore, the paper establishes a tight complexity dichotomy by showing tractability for P₅-free instances and intractability for P₁₀-free instances, extending these results to several related variants.
0 citationsRead paper
Multi-Modal Machine Learning for Population- and Subject-Specific lncRNA-Type 2 Diabetes Association Analysis
Traditional single-omics approaches struggle to fully elucidate the complex regulatory mechanisms of long non-coding RNAs (lncRNAs) in type 2 diabetes (T2D). This study proposes the first multimodal machine learning framework integrating lncRNA expression profiles, secondary structures, and sequence features. By combining eight classifiers, hierarchical cross-validation, and SHAP-based interpretability analysis, the framework enables precise association mapping from population to individual levels across two independent cohorts. The approach identifies multiple T2D-significant lncRNAs—including GAS5, XIST, MEG3, and ANRIL—with MEG3 consistently emerging as the top cross-cohort driver according to SHAP values. These findings not only corroborate results from conventional statistical methods but also yield a higher-resolution regulatory landscape, advancing the potential of lncRNAs in T2D precision medicine.
0 citationsRead paper
PhaseNet++: Phase-Aware Frequency-Domain Anomaly Detection for Industrial Control Systems via Phase Coherence Graphs
This work addresses the common oversight of phase information in the frequency domain for multivariate time series anomaly detection in industrial control systems, which often hinders the identification of synchronized anomalies across sensors. To this end, the authors propose PhaseNet++, a novel framework that systematically incorporates phase cues by extracting magnitude and phase spectra via short-time Fourier transform, constructing a lightweight graph structure using phase coherence indices, and jointly reconstructing signals through an integrated architecture combining graph attention networks and a Sensor-Token Transformer. A dual-head decoder—featuring recurrent and coherence-aware branches—is further introduced to enhance modeling capacity. Evaluated on the SWaT dataset, PhaseNet++ achieves an F1-score of 90.98%, ROC-AUC of 95.66%, and average precision of 91.51%, substantially outperforming existing methods with only 264,816 additional parameters.
0 citationsRead paper
ioPUF+: A PUF Based on I/O Pull-Up/Down Resistors for Secret Key Generation in IoT Nodes
To address the demand for low-cost, off-the-shelf hardware–compatible Physical Unclonable Functions (PUFs) in IoT node key generation, this paper proposes a lightweight PUF mechanism leveraging process variations in commercial chips’ I/O pull-up/pull-down resistors. The approach requires no IC design or fabrication modifications and extracts device-unique fingerprints solely from standard I/O structures. It enables end-to-end conversion from raw responses to cryptographically secure keys via resistance measurement, BCH error correction, and SHA-256 hashing, integrated within an AES-secured communication pipeline. Evaluated on 30 commercial MCUs, the PUF achieves 100% intra-chip Hamming distance consistency, 50.33% inter-chip uniqueness, 50.54% response uniformity, and a worst-case bit error rate <2.63%. The implementation occupies only 19.8 KB of flash memory, consumes 79 mW, and incurs a key generation latency of 600 ms.
0 citationsRead paper