Stony Brook University Computer Science

Prof. Zhoulai Fu has received a Starting Program Grant from the National Research Foundation of Korea to support his research in Numerical Software. His work focuses on developing reliable and efficient numerical software using advanced programming languages and software engineering solutions. The grant provides funding from  late 2022 to early 2025.

Numerical software is used across various fields, including artificial intelligence, engineering, robotics, and medical imaging, but it can produce misleading or erroneous results if not built with care. Professor Fu's project aims to address this challenge by building trustworthy and performant numerical software with formal methods.

As part of this effort, he is investigating the use of pure 16-bit neural networks over 32-bit for image classification problems. Image classification involves automatically categorizing images into predefined categories and is an important problem in machine learning and everyday life, such as in facial recognition. Professor Fu's current work seeks to explore the surprising performance benefits of using 16-bit neural networks over the standard, widely used 32-bit neural networks.