About Me
Hello and welcome to my website dedicated to the DREAM project. My name is Kexin Hao, but you can call me Heather. I am currently pursuing a master’s degree in computer science in the Align program at Northeastern University in Seattle, with an expected graduation date in May 2025. Prior to enrolling in the NEU Align program, I completed my B.A. degree in chemistry at Franklin and Marshall College and was a candidate for a VMD degree at the UPenn School of Veterinary Medicine. During that time, my researches focused primarily on animals and chemicals, and I also used some technologies related to computer vision. However, due to personal health and financial reasons, I had to change career paths. It was during the pandemic that my interest in computer science was sparked when my husband proposed to me using an animation he created with Turtle in Python. Prior to joining NEU, I took the initiative to design my own personal CV-site completely from scratch. You can visit the site here. Additionally, a user friendly version personal website can be found here.
After completing a year of training at NEU Align, I am thrilled to have been selected for the Distributed Research Apprenticeships for Master’s (DREAM) program. The DREAM program is specifically designed to provide research experience and support to students like myself who do not hold a bachelor’s degree in computer science, as we pursue higher education in computing. I am incredibly grateful for the opportunity to delve deeper into the field of computer vision through this program and to receive valuable mentorship.
Outside of my studies, I find joy in playing tennis and spending quality time with my beloved dogs and cats.
If you have any questions or if you would like to collaborate, please don’t hesitate to reach out to me using the email link at the bottom of the page.
About My Advisor
Teaching professor and assistant director for computing programs at NEU Seattle
Research Interests: Computer vision, Machine learning, Robotics, Data science
About My Project
The project will focus on exploring the impact of using log RGB images in the task of object recognition. It has been recognized in the previous studies that the typical signal processing pipeline and compression methods, such as JPEG, could break the physical rules governing the interaction of light and matter, making color and intensity unreliable features for learning. By employing linear or log RGB images that preserve the physics of reflection, we aim to simplify the learning process for object recognition tasks and enhance robustness to certain types of visual variations.
To conduct our experiment, we will start with a pretrained network for object recognition and fine-tune it using log image data. By leveraging the log RGB images, we anticipate that the training process will be improved and lead to better performance. The goal is to assess whether the utilization of log data can effectively enhance the network’s ability to learn and generalize object recognition tasks. Through this investigation, we hope to validate the hypothesis that incorporating log RGB images can enable the use of a lighter weight network while maintaining or improving performance in object recognition.
My Blog
Please review my blogs for a comprehensive update on the progress of my research.