Task-Specific Prompt with Global Context for Multi-Task Graph Pre-Training
为解决图预训练中任务相关性和结构感知性弱的问题，提出TPGC方法，通过任务先验和结构先验协同初始化提示，提升少样本场景下的性能。
0 citationsRead paper
Selected work
Representative Papers
Secure Coverage Enhancement in Aerial Reconfigurable Intelligent Surface-Assisted High-Speed Train Communication Systems
This work addresses the challenge of ensuring physical-layer security in high-speed train communications, where achieving high reliability, high data rates, and resilience against eavesdropping simultaneously remains difficult. To this end, it introduces aerial reconfigurable intelligent surfaces (ARIS) into this scenario for the first time and proposes a joint optimization framework that coordinates base station active beamforming with ARIS phase shifts to maximize the weighted sum secrecy rate, subject to transmit power and unit-modulus constraints. The problem is decomposed via block coordinate descent; successive convex approximation is employed to optimize beamforming, while the alternating direction method of multipliers efficiently updates the ARIS phase shifts. The proposed algorithm converges rapidly, and simulations demonstrate its significant performance gains over existing approaches, effectively enhancing the system’s secrecy performance.
0 citationsRead paper
BAN: Neuroanatomical Aligning in Auditory Recognition between Artificial Neural Network and Human Cortex
Conventional artificial neural networks (ANNs) achieve high performance in auditory recognition but lack biological plausibility—particularly due to excessive depth and absence of recurrent connectivity—hindering alignment with the human auditory pathway. Method: We propose the Brain-inspired Auditory Network (BAN), the first ANN architecture explicitly mapped to the neuroanatomical structure of human temporal lobe auditory cortices (T2/T3), incorporating recurrent connections to enhance biological fidelity. We introduce the Brain-like Auditory Score (BAS), a novel cross-species metric quantifying functional similarity between artificial and biological auditory systems. Contribution/Results: Validated through neuroanatomical modeling, multi-scale functional alignment assessment, and music genre classification, BAN significantly outperforms deep ANNs in both recognition accuracy and BAS. Results demonstrate a strong correlation between auditory recognition capability and structural similarity to cortical auditory organization, establishing neuroanatomical alignment as a key design principle for biologically grounded auditory AI.
0 citationsRead paper
Recent publications
Latest Papers
Task-Specific Prompt with Global Context for Multi-Task Graph Pre-Training
为解决图预训练中任务相关性和结构感知性弱的问题，提出TPGC方法，通过任务先验和结构先验协同初始化提示，提升少样本场景下的性能。
0 citationsRead paper
Secure Coverage Enhancement in Aerial Reconfigurable Intelligent Surface-Assisted High-Speed Train Communication Systems
This work addresses the challenge of ensuring physical-layer security in high-speed train communications, where achieving high reliability, high data rates, and resilience against eavesdropping simultaneously remains difficult. To this end, it introduces aerial reconfigurable intelligent surfaces (ARIS) into this scenario for the first time and proposes a joint optimization framework that coordinates base station active beamforming with ARIS phase shifts to maximize the weighted sum secrecy rate, subject to transmit power and unit-modulus constraints. The problem is decomposed via block coordinate descent; successive convex approximation is employed to optimize beamforming, while the alternating direction method of multipliers efficiently updates the ARIS phase shifts. The proposed algorithm converges rapidly, and simulations demonstrate its significant performance gains over existing approaches, effectively enhancing the system’s secrecy performance.
0 citationsRead paper
BAN: Neuroanatomical Aligning in Auditory Recognition between Artificial Neural Network and Human Cortex
Conventional artificial neural networks (ANNs) achieve high performance in auditory recognition but lack biological plausibility—particularly due to excessive depth and absence of recurrent connectivity—hindering alignment with the human auditory pathway. Method: We propose the Brain-inspired Auditory Network (BAN), the first ANN architecture explicitly mapped to the neuroanatomical structure of human temporal lobe auditory cortices (T2/T3), incorporating recurrent connections to enhance biological fidelity. We introduce the Brain-like Auditory Score (BAS), a novel cross-species metric quantifying functional similarity between artificial and biological auditory systems. Contribution/Results: Validated through neuroanatomical modeling, multi-scale functional alignment assessment, and music genre classification, BAN significantly outperforms deep ANNs in both recognition accuracy and BAS. Results demonstrate a strong correlation between auditory recognition capability and structural similarity to cortical auditory organization, establishing neuroanatomical alignment as a key design principle for biologically grounded auditory AI.
0 citationsRead paper