陈伦 Chen Lun

Purpose: Deep learning has been widely used in medical image analysis. However, deep learning models are vulnerable to adversarial attacks, which could be a potential security threat to medical image analysis systems. Methods: In this study, we proposed a novel defense method for medical image analysis systems against adversarial attacks. Our method combines network pruning and attention mechanism to enhance the robustness of deep learning models. We evaluated our method on three medical imaging datasets under various adversarial attacks. Results: Experimental results show that our proposed method can effectively defend against adversarial attacks while maintaining high accuracy on clean images. Conclusions: Our work provides a promising approach to enhance the security of medical image analysis systems in clinical applications.

Tactile perception plays a crucial role in robotic manipulation tasks. However, effectively processing and interpreting tactile data remains challenging. In this paper, we propose Tactile-GAT, a novel graph attention network for tactile perception classification. Our approach models tactile sensor data as graphs and leverages attention mechanisms to capture spatial relationships between tactile elements. Experimental results demonstrate that our method achieves superior performance compared to traditional approaches, enabling more accurate object recognition and material classification through tactile sensing.