Selected work

Representative Papers

From Neural Intent to Cryptographic Authorization: Governing Agentic Workflows

Jul 16, 2026

This work addresses the vulnerability of existing key management services to prompt injection attacks, stemming from their inability to verify alignment between AI agent runtime behavior and user intent. To mitigate this, the paper introduces Neural Cryptographic Service (NCS), the first framework integrating cryptographic authorization with neuro-symbolic control. In NCS, a neural planner generates an initial action plan, while a symbolic controller enforces parameter binding, fine-grained authorization, and tamper-proof execution for each tool invocation, leveraging offline digital signatures and hash chains. This approach enables auditable, deterministic runtime governance, reducing attack success rates to near zero on both AgentDojo and custom benchmarks while preserving high availability for legitimate tasks.

0 citationsRead paper

A Large-Scale Sparse Multiobjective Optimization Algorithm Based on Optimal Performance Scores

Jul 06, 2026

This work addresses the performance degradation in large-scale sparse multi-objective optimization caused by the difficulty of accurately identifying high-dimensional variables and critical non-zero components. To tackle this challenge, a novel evolutionary algorithm is proposed that leverages an optimal performance score–driven initialization and an initial mask template to locate key variables, while employing a Pareto-guided normal distribution to optimize real-valued dimensions. The method innovatively integrates a variable importance scoring mechanism and an adaptive mutation probability strategy, substantially enhancing the accuracy of sparse solution identification and overall algorithmic robustness. Extensive experiments on eight benchmark problems and three real-world applications demonstrate that the proposed approach significantly outperforms state-of-the-art algorithms in both convergence speed and solution accuracy.

0 citationsRead paper

High-Precision Formation Control for Heterogeneous Multi-Robot Systems via Hierarchical Hybrid Physics-Informed Deep Reinforcement Learning

Jul 03, 2026

This work addresses the limitations of traditional model-based control—its reliance on precise dynamics and poor handling of uncertainty—and the low sample efficiency and poor convergence often observed in end-to-end deep reinforcement learning (DRL). To overcome these challenges, the paper proposes a Hierarchical Hybrid Physics-Informed Deep Reinforcement Learning framework (HHy-PIDRL). The upper layer employs Soft Actor-Critic (SAC) to generate leader navigation policies, while the lower layer integrates high-fidelity physics-based feedforward control, PD feedback, and an adaptive DRL residual controller, establishing a synergistic model-learning paradigm for formation control. A hierarchical reward function is designed to train omnidirectional follower robots. Experimental results demonstrate 100% success rates in both navigation and formation tasks, and ablation studies confirm the proposed architecture’s superior accuracy, responsiveness, and robustness.

0 citationsRead paper

LRX-PINN: A Layer-Resolving XNet Physics-Informed Neural Network with Integrated Cauchy Activations for Convection-Dominated Problems

Jul 03, 2026

This work addresses the challenge that standard physics-informed neural networks (PINNs) struggle to accurately approximate solutions and their derivatives in convection-dominated problems featuring thin-layer structures. To overcome this, the authors propose LRX-PINN, a method that constructs basis functions with transition-like behavior by incorporating Cauchy activation functions. These basis functions exhibit transitional profiles in the solution space, while their derivatives correspond to localized Cauchy kernels that precisely align with the scale of thin layers. The resulting neural representation is structurally consistent with the underlying solution, offering both derivative stability and exponential approximation capability. The study further reveals that the effective physical width of ridge neurons is $d/|w|$. Implemented within an XNet architecture and embedded into an hp-VPINN framework, LRX-PINN achieves higher accuracy than existing methods using fewer than 30% of their parameters, and further surpasses state-of-the-art performance when integrated into the hp-VPINN framework.

0 citationsRead paper

ID-Eraser: Proactive Defense Against Face Swapping via Identity Perturbation

Apr 23, 2026

This work addresses the growing threat of deepfake-based face swapping to privacy and security, noting that existing pixel-level defenses are ineffective against advanced models leveraging high-dimensional identity embeddings. To counter this, the authors propose the first approach that introduces learnable perturbations directly in the identity embedding space to actively erase identifiable information, while employing a face restoration generator to produce visually natural images. Evaluated under black-box settings, the method drastically reduces the adversary’s ability to exploit identity cues—lowering Top-1 identification accuracy to 0.30 and mean identity similarity to 0.504—without compromising perceptual quality, as evidenced by a low FID of 1.64 and LPIPS of 0.020. The approach further demonstrates strong generalization and robustness across datasets and common image distortions.

0 citationsRead paper

Recent publications

Latest Papers

From Neural Intent to Cryptographic Authorization: Governing Agentic Workflows

Jul 16, 2026

This work addresses the vulnerability of existing key management services to prompt injection attacks, stemming from their inability to verify alignment between AI agent runtime behavior and user intent. To mitigate this, the paper introduces Neural Cryptographic Service (NCS), the first framework integrating cryptographic authorization with neuro-symbolic control. In NCS, a neural planner generates an initial action plan, while a symbolic controller enforces parameter binding, fine-grained authorization, and tamper-proof execution for each tool invocation, leveraging offline digital signatures and hash chains. This approach enables auditable, deterministic runtime governance, reducing attack success rates to near zero on both AgentDojo and custom benchmarks while preserving high availability for legitimate tasks.

0 citationsRead paper

A Large-Scale Sparse Multiobjective Optimization Algorithm Based on Optimal Performance Scores

Jul 06, 2026

This work addresses the performance degradation in large-scale sparse multi-objective optimization caused by the difficulty of accurately identifying high-dimensional variables and critical non-zero components. To tackle this challenge, a novel evolutionary algorithm is proposed that leverages an optimal performance score–driven initialization and an initial mask template to locate key variables, while employing a Pareto-guided normal distribution to optimize real-valued dimensions. The method innovatively integrates a variable importance scoring mechanism and an adaptive mutation probability strategy, substantially enhancing the accuracy of sparse solution identification and overall algorithmic robustness. Extensive experiments on eight benchmark problems and three real-world applications demonstrate that the proposed approach significantly outperforms state-of-the-art algorithms in both convergence speed and solution accuracy.

0 citationsRead paper

High-Precision Formation Control for Heterogeneous Multi-Robot Systems via Hierarchical Hybrid Physics-Informed Deep Reinforcement Learning

Jul 03, 2026

This work addresses the limitations of traditional model-based control—its reliance on precise dynamics and poor handling of uncertainty—and the low sample efficiency and poor convergence often observed in end-to-end deep reinforcement learning (DRL). To overcome these challenges, the paper proposes a Hierarchical Hybrid Physics-Informed Deep Reinforcement Learning framework (HHy-PIDRL). The upper layer employs Soft Actor-Critic (SAC) to generate leader navigation policies, while the lower layer integrates high-fidelity physics-based feedforward control, PD feedback, and an adaptive DRL residual controller, establishing a synergistic model-learning paradigm for formation control. A hierarchical reward function is designed to train omnidirectional follower robots. Experimental results demonstrate 100% success rates in both navigation and formation tasks, and ablation studies confirm the proposed architecture’s superior accuracy, responsiveness, and robustness.

0 citationsRead paper

LRX-PINN: A Layer-Resolving XNet Physics-Informed Neural Network with Integrated Cauchy Activations for Convection-Dominated Problems

Jul 03, 2026

This work addresses the challenge that standard physics-informed neural networks (PINNs) struggle to accurately approximate solutions and their derivatives in convection-dominated problems featuring thin-layer structures. To overcome this, the authors propose LRX-PINN, a method that constructs basis functions with transition-like behavior by incorporating Cauchy activation functions. These basis functions exhibit transitional profiles in the solution space, while their derivatives correspond to localized Cauchy kernels that precisely align with the scale of thin layers. The resulting neural representation is structurally consistent with the underlying solution, offering both derivative stability and exponential approximation capability. The study further reveals that the effective physical width of ridge neurons is $d/|w|$. Implemented within an XNet architecture and embedded into an hp-VPINN framework, LRX-PINN achieves higher accuracy than existing methods using fewer than 30% of their parameters, and further surpasses state-of-the-art performance when integrated into the hp-VPINN framework.

0 citationsRead paper

ID-Eraser: Proactive Defense Against Face Swapping via Identity Perturbation

Apr 23, 2026

This work addresses the growing threat of deepfake-based face swapping to privacy and security, noting that existing pixel-level defenses are ineffective against advanced models leveraging high-dimensional identity embeddings. To counter this, the authors propose the first approach that introduces learnable perturbations directly in the identity embedding space to actively erase identifiable information, while employing a face restoration generator to produce visually natural images. Evaluated under black-box settings, the method drastically reduces the adversary’s ability to exploit identity cues—lowering Top-1 identification accuracy to 0.30 and mean identity similarity to 0.504—without compromising perceptual quality, as evidenced by a low FID of 1.64 and LPIPS of 0.020. The approach further demonstrates strong generalization and robustness across datasets and common image distortions.

0 citationsRead paper