Songtao Lu

Title: Songtao Lu: Innovator in Machine Learning

Introduction

Songtao Lu is a prominent inventor based in White Plains, NY (US). He has made significant contributions to the field of machine learning, particularly in the area of adversarial training techniques. With a total of 4 patents, his work focuses on enhancing the robustness of machine learning models.

Latest Patents

One of his latest patents is titled "Generating unsupervised adversarial examples for machine learning." This invention involves receiving a trained machine learning model and a training dataset used to train that model. Based on the training dataset, unsupervised adversarial examples can be generated. The robustness of the trained machine learning model can be determined using these generated examples. Additionally, the training dataset can be augmented with the unsupervised adversarial examples, allowing for retraining of the model with the enhanced dataset.

Another notable patent is "Distributed adversarial training for robust deep neural networks." This invention provides scalable distributed adversarial training techniques for deep neural networks. The method involves obtaining adversarial perturbation-modified training examples and computing gradients of a local cost function with respect to the model's parameters. These gradients are then transmitted to a server for aggregation, which updates the model parameters stored at each distributed computing machine.

Career Highlights

Songtao Lu has worked with notable organizations, including IBM and National Chung Hsing University. His experience in these institutions has allowed him to collaborate with leading experts in the field of machine learning.

Collaborations

Some of his coworkers include Sijia Liu and Chuang Gan, who have also contributed to advancements in machine learning.

Conclusion

Songtao Lu's innovative work in machine learning, particularly in adversarial training, showcases his commitment to enhancing the capabilities of artificial intelligence. His patents reflect a deep understanding of the challenges in the field and offer solutions that can significantly improve model robustness.