Wang Zhang

Title: Innovations of Wang Zhang in Neural Network Robustness

Introduction

Wang Zhang is an accomplished inventor based in Cambridge, MA (US). He has made significant contributions to the field of neural networks, particularly in enhancing the robustness of image classification systems. His innovative approach focuses on certifying the reliability of neural network models.

Latest Patents

Wang Zhang holds a patent titled "Certification-based robust training by refining decision boundary." This patent describes a computer-implemented method for certifying the robustness of image classification in a neural network. The method involves initializing a neural network model that includes a problem space and a decision boundary. A processor receives a dataset of images, image labels, and a perturbation schedule. Images are drawn from the dataset in the problem space, and a distance from the decision boundary is determined for these images. A re-weighting value is applied, along with a modified perturbation magnitude. The total loss function for the images is calculated using the re-weighting value, and a confidence level of the classification is evaluated for certifiable robustness.

Career Highlights

Throughout his career, Wang Zhang has worked with prestigious organizations, including IBM and the Massachusetts Institute of Technology. His experience in these institutions has allowed him to refine his skills and contribute to groundbreaking research in neural networks.

Collaborations

Wang has collaborated with notable colleagues such as Alexandre Megretski and Lam Nguyen. Their joint efforts have further advanced the field of neural network robustness.

Conclusion

Wang Zhang's innovative work in certifying the robustness of neural networks has made a significant impact in the field of image classification. His contributions continue to influence advancements in technology and research.