Selected work

Representative Papers

Communication-Aware Multi-Agent Reinforcement Learning for Decentralized Cooperative UAV Deployment

Mar 17, 2026

This work addresses the challenge of cooperative deployment for drone swarms under partial observability and intermittent communication by proposing a graph neural network–based multi-agent reinforcement learning approach grounded in the centralized training with decentralized execution (CTDE) paradigm. The method employs a distance-constrained communication graph and introduces agent-entity attention alongside neighbor self-attention mechanisms, enabling efficient coordination using only local observations and messages from nearby agents. This architecture supports zero-shot generalization to formations of varying scales. Experimental results demonstrate that, in the DroneConnect task, a team of five drones achieves 74% area coverage—approaching the offline upper bound provided by mixed-integer linear programming—and significantly outperforms non-communicating baselines in the DroneCombat task, thereby validating the approach’s effectiveness and scalability.

0 citationsRead paper

Uncovering the Fragility of Trustworthy LLMs through Chinese Textual Ambiguity

Jul 30, 2025

This study exposes critical trustworthiness deficiencies in large language models (LLMs) regarding Chinese textual ambiguity understanding: LLMs frequently fail to detect ambiguity, exhibit overconfidence in single interpretations, and generate redundant reasoning during polysemy resolution. To address this, the authors introduce the first context-augmented benchmark dataset specifically designed for Chinese ambiguity understanding—covering three major categories and nine fine-grained ambiguity types—curated and annotated manually to ensure high quality. They further propose a multidimensional evaluation framework to systematically assess ambiguity detection, sense discrimination, and confidence calibration capabilities. Experimental results demonstrate that state-of-the-art LLMs significantly underperform humans in ambiguity awareness and uncertainty handling, revealing fundamental limitations in their cognitive modeling. The dataset and evaluation code are publicly released, establishing essential infrastructure for trustworthy NLP research.

0 citationsRead paper

Quantum Dynamics Simulation of the Advection-Diffusion Equation

Mar 17, 2025

Solving convection–diffusion partial differential equations (PDEs) on quantum hardware remains challenging due to qubit overhead and noise sensitivity. Method: We present the first auxiliary-qubit-free quantum simulation of the convection–diffusion equation on a superconducting quantum platform. Finite-difference operators are directly encoded into the Hamiltonian, and three algorithms—Trotterization, variational quantum time evolution (VarQTE), and adaptive variational quantum dynamical simulation (AVQDS)—are implemented and benchmarked on both the IBM Qiskit Aer simulator and the Fez superconducting quantum processor. Contribution/Results: On the simulator, quantum solutions achieve fidelity ∼1−10⁻⁵ against direct numerical solutions (DNS). AVQDS minimizes gate count and circuit depth, while Trotterization attains the highest accuracy. Hardware experiments quantify error bounds under current noise levels. This work establishes a novel paradigm for adapting many-body quantum dynamical algorithms to classical transport PDEs and provides a scalable pathway toward quantum-accelerated computational fluid dynamics.

0 citationsRead paper

Recent publications

Latest Papers

Communication-Aware Multi-Agent Reinforcement Learning for Decentralized Cooperative UAV Deployment

Mar 17, 2026

This work addresses the challenge of cooperative deployment for drone swarms under partial observability and intermittent communication by proposing a graph neural network–based multi-agent reinforcement learning approach grounded in the centralized training with decentralized execution (CTDE) paradigm. The method employs a distance-constrained communication graph and introduces agent-entity attention alongside neighbor self-attention mechanisms, enabling efficient coordination using only local observations and messages from nearby agents. This architecture supports zero-shot generalization to formations of varying scales. Experimental results demonstrate that, in the DroneConnect task, a team of five drones achieves 74% area coverage—approaching the offline upper bound provided by mixed-integer linear programming—and significantly outperforms non-communicating baselines in the DroneCombat task, thereby validating the approach’s effectiveness and scalability.

0 citationsRead paper

Uncovering the Fragility of Trustworthy LLMs through Chinese Textual Ambiguity

Jul 30, 2025

This study exposes critical trustworthiness deficiencies in large language models (LLMs) regarding Chinese textual ambiguity understanding: LLMs frequently fail to detect ambiguity, exhibit overconfidence in single interpretations, and generate redundant reasoning during polysemy resolution. To address this, the authors introduce the first context-augmented benchmark dataset specifically designed for Chinese ambiguity understanding—covering three major categories and nine fine-grained ambiguity types—curated and annotated manually to ensure high quality. They further propose a multidimensional evaluation framework to systematically assess ambiguity detection, sense discrimination, and confidence calibration capabilities. Experimental results demonstrate that state-of-the-art LLMs significantly underperform humans in ambiguity awareness and uncertainty handling, revealing fundamental limitations in their cognitive modeling. The dataset and evaluation code are publicly released, establishing essential infrastructure for trustworthy NLP research.

0 citationsRead paper

Quantum Dynamics Simulation of the Advection-Diffusion Equation

Mar 17, 2025

Solving convection–diffusion partial differential equations (PDEs) on quantum hardware remains challenging due to qubit overhead and noise sensitivity. Method: We present the first auxiliary-qubit-free quantum simulation of the convection–diffusion equation on a superconducting quantum platform. Finite-difference operators are directly encoded into the Hamiltonian, and three algorithms—Trotterization, variational quantum time evolution (VarQTE), and adaptive variational quantum dynamical simulation (AVQDS)—are implemented and benchmarked on both the IBM Qiskit Aer simulator and the Fez superconducting quantum processor. Contribution/Results: On the simulator, quantum solutions achieve fidelity ∼1−10⁻⁵ against direct numerical solutions (DNS). AVQDS minimizes gate count and circuit depth, while Trotterization attains the highest accuracy. Hardware experiments quantify error bounds under current noise levels. This work establishes a novel paradigm for adapting many-body quantum dynamical algorithms to classical transport PDEs and provides a scalable pathway toward quantum-accelerated computational fluid dynamics.

0 citationsRead paper