Institution profile

Taiyuan University of Technology

Academic institutionasia · cn
Official website
Research library14linked papers
Opportunities0open roles
Selected work

Representative Papers

Gaussian-JEPA: Joint-Embedding Predictive Learning for 3D Gaussian Splats

Aug 16, 2026

This study addresses the poor generalization in 3D Gaussian self-supervised learning caused by attribute coupling and proposes the Gaussian-JEPA framework. By replacing pixel-level reconstruction with latent space prediction, this method leverages a joint embedding architecture and multi-scale target generation to decouple geometric and appearance supervision, enabling efficient decoder-free representation learning. Experiments demonstrate that the model yields more consistent representations under resampling and partial observation conditions. Furthermore, frozen features from Gaussian-JEPA significantly outperform those from traditional reconstruction-based pretraining across downstream tasks, including shape completion, part segmentation, and classification. These results confirm that the proposed approach effectively enhances the versatility and reusability of 3D Gaussian representations, establishing a robust foundation for diverse 3D vision applications without requiring task-specific fine-tuning of the backbone encoder.

0 citationsRead paper

AmbSentry: Mitigating Sensing Eavesdropping in ISAC Systems by Harnessing Ambient IoT Devices

Aug 12, 2026

This work addresses the vulnerability of sensing information leakage in integrated sensing and communication (ISAC) systems due to the open nature of wireless channels. It proposes, for the first time, leveraging passive ambient Internet-of-Things (AIoT) devices as cooperative jammers and virtual targets to actively establish a physical-layer sensing security mechanism. By jointly optimizing base station transmit beamforming and AIoT reflection modulation, the approach injects controlled interference without relying on conventional encryption, thereby preserving sensing and communication performance for legitimate users while degrading an eavesdropper’s ability to estimate target parameters. An efficient algorithm based on the Dinkelbach transformation and block coordinate descent is developed to solve the resulting non-convex optimization problem. Experimental results demonstrate that the legitimate receiver achieves a 14 dB SNR gain in detection probability and exhibits parameter estimation errors two orders of magnitude lower than those of the eavesdropper, significantly enhancing sensing privacy and security.

0 citationsRead paper
Recent publications

Latest Papers

Gaussian-JEPA: Joint-Embedding Predictive Learning for 3D Gaussian Splats

Aug 16, 2026

This study addresses the poor generalization in 3D Gaussian self-supervised learning caused by attribute coupling and proposes the Gaussian-JEPA framework. By replacing pixel-level reconstruction with latent space prediction, this method leverages a joint embedding architecture and multi-scale target generation to decouple geometric and appearance supervision, enabling efficient decoder-free representation learning. Experiments demonstrate that the model yields more consistent representations under resampling and partial observation conditions. Furthermore, frozen features from Gaussian-JEPA significantly outperform those from traditional reconstruction-based pretraining across downstream tasks, including shape completion, part segmentation, and classification. These results confirm that the proposed approach effectively enhances the versatility and reusability of 3D Gaussian representations, establishing a robust foundation for diverse 3D vision applications without requiring task-specific fine-tuning of the backbone encoder.

0 citationsRead paper

AmbSentry: Mitigating Sensing Eavesdropping in ISAC Systems by Harnessing Ambient IoT Devices

Aug 12, 2026

This work addresses the vulnerability of sensing information leakage in integrated sensing and communication (ISAC) systems due to the open nature of wireless channels. It proposes, for the first time, leveraging passive ambient Internet-of-Things (AIoT) devices as cooperative jammers and virtual targets to actively establish a physical-layer sensing security mechanism. By jointly optimizing base station transmit beamforming and AIoT reflection modulation, the approach injects controlled interference without relying on conventional encryption, thereby preserving sensing and communication performance for legitimate users while degrading an eavesdropper’s ability to estimate target parameters. An efficient algorithm based on the Dinkelbach transformation and block coordinate descent is developed to solve the resulting non-convex optimization problem. Experimental results demonstrate that the legitimate receiver achieves a 14 dB SNR gain in detection probability and exhibits parameter estimation errors two orders of magnitude lower than those of the eavesdropper, significantly enhancing sensing privacy and security.

0 citationsRead paper