Summer 2025 @ Oak Ridge National Laboratory
During the summer of 2025, I interned at Oak Ridge National Laboratory (ORNL). Over the 10 weeks, I worked on porting and running computational pathology models on Frontier, which is currently one of the fastest supercomputers in the world.
Project
My main goal was to take existing deep learning models used for cancer classification—specifically TransMIL and CLAM—and get them running.
Orginally, these existing codebases were built for NVIDIA GPUs using CUDA, but Frontier runs on AMD hardware. That meant I had to do quite a bit of work to get everything compatible with ROCm so it could properly run on Frontier.
Pipeline & Datasets
Once the framework was set up, I worked on the full end-to-end machine learning pipeline, which included:
- Data Preprocessing: Wrote distributed data preprocessing scripts to handle large-scale histopathology datasets which averages 30,000 patches at 10x magnification emulating what a pathologist would see under a microscope. each patch is 256x256
- Datasets: Tested and Benchmarked the pipeline using LUAD, LUSC, and DLBC datasets.
- Embedding and Training: Generated the image embeddings, trained the models, and ran evaluations using 10-fold cross-validation.
2025 Smoky Mountains Computational Sciences and Engineering Conference (SMC 25)
At the end of the internship, I was invited as one of the few student poster presenters to the 2025 SMC conference to showcase the work. It was an incredible experience to meet and network with other students, researchers, and professionals in the field of computational sciences and engineering.