I am currently working as a research assistant with the Distributed Protocols Research Group, at UIUC.
- Designed fault-tolerant distributed ML training platforms for model and data parallelism across GPUs, minimizing training time for transformer-like deep learning models.
- Enhanced functionality of internal memory allocators to analyze advantages of custom-placed operators (step time and memory), as opposed to Tensorflow’s native placement.
- Conducted computational experiments to evaluate Tensorflow memory allocation at varying transformer model sizes.
Working alongside two other undergraduate students, I developed a program to effectively analyze images and determine the skin color of
lead actors - this would be used to perform social critiques of film.
- Developed scripts to detect faces within moving and still images through computer vision algorithms in Google MediaPipe and OpenCV2.
- Implemented non-negative matrix factorization and principal component analysis to extract skin colors from facial patches, which were used to calculate average skin color.
- Wrote multithreaded Python webscrapers to generate testing/training datasets by batch-downloading sets of 1000+ images.