Zhi-Gang Liu

Title: Zhi-Gang Liu: Innovator in Deep Learning Acceleration

Introduction

Zhi-Gang Liu is a prominent inventor based in Westford, MA (US), known for his significant contributions to the field of deep learning and integrated circuit design. With a total of 16 patents to his name, Liu has made remarkable advancements that enhance computational efficiency and performance.

Latest Patents

Among his latest innovations, Liu has developed an activation compression method for deep learning acceleration. This system and method facilitate the multiplication of matrices, which is crucial for various machine learning applications. The system includes a processor and a matrix multiply accelerator (MMA). The processor generates basic block matrices from an input tensor and prunes elements based on a sparsity value. It also creates a mask for each basic block matrix, compressing it to produce a more efficient output. The MMA then utilizes these masks to multiply the compressed matrices with a weight matrix, resulting in an output matrix. Another notable patent is the CPU tight-coupled accelerator, which integrates a central processing unit (CPU) core with an accelerator and an acceleration instruction queue. This design allows for efficient instruction fetching and execution, particularly for acceleration workloads.

Career Highlights

Liu has worked with several notable companies throughout his career, including Arm Limited and EMC Corporation. His experience in these organizations has contributed to his expertise in developing cutting-edge technologies in the field of computing.

Collaborations

Liu has collaborated with talented individuals such as Paul Nicholas Whatmough and Matthew Mattina, further enhancing his innovative projects and research.

Conclusion

Zhi-Gang Liu's work in deep learning acceleration and integrated circuit design showcases his dedication to advancing technology. His patents reflect a commitment to improving computational methods, making significant impacts in the industry.