Selected work

Representative Papers

Variational Quantum Attention for Molecular Graph Learning

Oct 03, 2026

This study addresses the limitations of classical attention mechanisms in molecular graph learning by investigating how variational quantum circuits can reshape attention behavior. We propose an edge-aware variational quantum attention mechanism that introduces a novel parameterization scheme encoding central atoms, neighboring atoms, and chemical bonds into quantum states, thereby enabling synergistic modeling of atomic and bond features. This mechanism reveals distinct attribution patterns between quantum and classical attention. Evaluated across five molecular property prediction tasks, our approach achieves performance comparable to GATv2, with particularly significant improvements on the BBBP dataset. Furthermore, a BACE1 case study validates its advantages in maintaining structure–activity relationship consistency.

0 citationsRead paper

Governed Deduction: Policy-Grounded Premise Authorization Beyond Relevance

Sep 25, 2026

This study addresses the limitation of existing reasoning systems that focus solely on premise relevance while neglecting policy authorization constraints. To this end, this work proposes a controlled deduction framework, constructs an RBAC-enhanced Spider benchmark, defines transfer local admission predicates, and introduces matching authorization pair construction, linear controllers, and context-local role permutation methods. Experimental results demonstrate that linear models fail to recover policy relations, with accuracy degrading to 50%, and exhibit limitations such as role-name shortcuts and frozen representations, whereas symbolic oracles maintain 100% accuracy. This research reveals the inherent deficiencies of linear models in permission reasoning and underscores the necessity of conducting leakage auditing for reasoning systems.

0 citationsRead paper

Zero-Shot Cross-Material Ptychographic Phase Reconstruction Using Deep Learning

Sep 12, 2026

该研究提出一种基于深度学习的直接局部到全局学习框架，用于从衍射测量中重建全视野相位图，无需迭代优化，解决了传统方法计算成本高的问题。

0 citationsRead paper

Hybrid Quantum-inspired Kolmogorov-Arnold Networks for Privacy-Aware Federated Biosignal Learning

Aug 13, 2026

This study addresses data heterogeneity, label imbalance, and communication bottlenecks in federated ECG classification by proposing a novel Hybrid Quantum-Inspired Kolmogorov-Arnold Network (QIKAN) integrated with the FedAvg framework. This approach effectively enhances robustness and parameter efficiency for cross-client arrhythmia classification while preserving privacy. Experimental results demonstrate that, compared to traditional MLPs, QIKAN reduces model parameters by 44.81% and communication overhead by 36.41%, while significantly improving classification metrics across most categories. Consequently, this work achieves efficient and precise distributed biosignal learning under strict privacy constraints, offering a promising solution for resource-constrained federated healthcare applications.

0 citationsRead paper

An Intelligent AI glasses System with Multi-Agent Architecture for Real-Time Voice Processing and Task Execution

Jan 09, 2026arXiv.org

This work addresses the challenges of achieving low-latency, multilingual voice interaction and cross-platform task automation for smart glasses in real-world scenarios. The authors propose an edge-oriented dual-agent collaborative architecture: Agent 01 handles multilingual speech recognition, while Agent 02 leverages a local large language model integrated with the MCP protocol, retrieval-augmented generation (RAG), and external tools to perform task reasoning and execution. The system supports RTSP audio-video streaming, eye-tracking data acquisition, and RabbitMQ-based remote communication, enabling end-to-end real-time voice command understanding and cross-platform task orchestration. Experimental results demonstrate the feasibility of deploying such a sophisticated AI agent system on resource-constrained wearable devices, significantly enhancing both interactive efficiency and multilingual adaptability.

0 citationsRead paper

Recent publications

Latest Papers

Variational Quantum Attention for Molecular Graph Learning

Oct 03, 2026

This study addresses the limitations of classical attention mechanisms in molecular graph learning by investigating how variational quantum circuits can reshape attention behavior. We propose an edge-aware variational quantum attention mechanism that introduces a novel parameterization scheme encoding central atoms, neighboring atoms, and chemical bonds into quantum states, thereby enabling synergistic modeling of atomic and bond features. This mechanism reveals distinct attribution patterns between quantum and classical attention. Evaluated across five molecular property prediction tasks, our approach achieves performance comparable to GATv2, with particularly significant improvements on the BBBP dataset. Furthermore, a BACE1 case study validates its advantages in maintaining structure–activity relationship consistency.

0 citationsRead paper

Governed Deduction: Policy-Grounded Premise Authorization Beyond Relevance

Sep 25, 2026

This study addresses the limitation of existing reasoning systems that focus solely on premise relevance while neglecting policy authorization constraints. To this end, this work proposes a controlled deduction framework, constructs an RBAC-enhanced Spider benchmark, defines transfer local admission predicates, and introduces matching authorization pair construction, linear controllers, and context-local role permutation methods. Experimental results demonstrate that linear models fail to recover policy relations, with accuracy degrading to 50%, and exhibit limitations such as role-name shortcuts and frozen representations, whereas symbolic oracles maintain 100% accuracy. This research reveals the inherent deficiencies of linear models in permission reasoning and underscores the necessity of conducting leakage auditing for reasoning systems.

0 citationsRead paper

Zero-Shot Cross-Material Ptychographic Phase Reconstruction Using Deep Learning

Sep 12, 2026

该研究提出一种基于深度学习的直接局部到全局学习框架，用于从衍射测量中重建全视野相位图，无需迭代优化，解决了传统方法计算成本高的问题。

0 citationsRead paper

Hybrid Quantum-inspired Kolmogorov-Arnold Networks for Privacy-Aware Federated Biosignal Learning

Aug 13, 2026

This study addresses data heterogeneity, label imbalance, and communication bottlenecks in federated ECG classification by proposing a novel Hybrid Quantum-Inspired Kolmogorov-Arnold Network (QIKAN) integrated with the FedAvg framework. This approach effectively enhances robustness and parameter efficiency for cross-client arrhythmia classification while preserving privacy. Experimental results demonstrate that, compared to traditional MLPs, QIKAN reduces model parameters by 44.81% and communication overhead by 36.41%, while significantly improving classification metrics across most categories. Consequently, this work achieves efficient and precise distributed biosignal learning under strict privacy constraints, offering a promising solution for resource-constrained federated healthcare applications.

0 citationsRead paper

An Intelligent AI glasses System with Multi-Agent Architecture for Real-Time Voice Processing and Task Execution

Jan 09, 2026arXiv.org

This work addresses the challenges of achieving low-latency, multilingual voice interaction and cross-platform task automation for smart glasses in real-world scenarios. The authors propose an edge-oriented dual-agent collaborative architecture: Agent 01 handles multilingual speech recognition, while Agent 02 leverages a local large language model integrated with the MCP protocol, retrieval-augmented generation (RAG), and external tools to perform task reasoning and execution. The system supports RTSP audio-video streaming, eye-tracking data acquisition, and RabbitMQ-based remote communication, enabling end-to-end real-time voice command understanding and cross-platform task orchestration. Experimental results demonstrate the feasibility of deploying such a sophisticated AI agent system on resource-constrained wearable devices, significantly enhancing both interactive efficiency and multilingual adaptability.

0 citationsRead paper