The following projects, ordered randomly, were carried out by the students in this class over a six-week period. Each student (or team) proposed a project, made changes based on the instructor's feedback on the proposal, presented an informal halfway talk, and gave a final presentation. Each project also produced a final report which are provided below. Some of the reports contain links to source code.
Johnny Li, AI-Guided LLVM Pass Ordering for Compute Kernels
Erik Wasmosy, Evaluating Quantization Methods for LLM Inference on Low-VRAM Consumer GPUs
Jeelin Liu and Minh Nguyen, Optimizing RAG Retrieval with Tiered Memory and Multi-Stage Indexing (Hashing + IVF + PQ)
Johnny Taveras, Optimizing Adversarial Defenses With CUDA
Jason Shin, Vectorization Additions to SMaT CPU Code and use Case Comparison of SMaT and dgl-SpMM
Lance Ulrich, Accelerating U-Net Inference for X-ray CT Defect Detection via Quantization and Compiler Optimization
Emilio Ochoa, Energy-Efficient Breast VSI Inference on Low-Cost Devices
Ivan Čabrilo, Scalable Training: Pipeline vs Data Parallelism for ResNet and Vision Transformers
Ethan Urmson and Amanda Jen, Quantizing Model Parameters for OCR Models
Siddharth Narsipur, OTX: Testing the ONNX Runtime Optimizer