Khushi D
- Research Program Mentor
MS candidate at Columbia University
Expertise
machine learning, deep learning, computer vision, artificial intelligence
Bio
I am Khushi Desai, a graduating Computer Science student at UC Berkeley and an incoming Masters of Computer Science student at Columbia University. I discovered my passion for Computer Vision and Deep Learning through college and have been working on research at Berkeley in the Autonomous Vehicles space for the last 2 years. I plan to transition to and pursue Computer Vision within the medical space over the course of my Masters degree. I am keen on exploring ways to leverage and optimize perception solutions that can ultimately contribute towards a better understanding and use of medical technology. My main hobbies include painting, playing musical instruments like piano and ukulele, and reading. I also enjoy outdoor activities like hiking, running, traveling, and scuba diving. Recently one of my coolest adventures was diving with giant manta rays in Kona Island, Hawaii!Project ideas
Research Paper on Generating Synthetic Data for Medical Imaging using NeRF
Computer vision within the medical field has become extremely relevant for identifying tumors, building more accurate diagnosis, estimating blood flow and blood pressure, etc. My expertise in computer vision is improving and optimizing computer generated scenes and semantic segmentation analyses of images. These tools can be applied to a range of topics, including autonomous vehicles, emotion recognition, medical radiology, etc. Through a project like this you will learn to understand existing technologies that are commonly used within the research industry (for example, NeRF), identify limitations of the technologies and propose improvements, research points of improvement, and build on top of the existing codebase. The potential outcomes of the project could involve writing a research paper by conducting experiments and building new code on top of existing codebases, or building a new application or project that leveraging existing technologies.
Image to Text Application Trained for Autonomous Vehicle Construction Sign Recognition
Computer vision is widely used for autonomous vehicle training and functionality. One of the applications is recognizing signs on the road and conveying information about the road signs to the car, in order to help it act accordingly and with caution. Construction signs are not often seen on the road and may require additional training in order to ensure that the car is ready to handle scenarios it has not seen before -- for example a sign with unseen writing. This can be done using computer vision and image to text machine learning algorithms. Through this project, you will learn to research and find existing models that perform the task of generating text from an image. You will be able to learn to modify, run, and embed these algorithms and models within your own encompassing application. The potential outcomes of this project would involve building an application that can be published through a GitHub repository.