Mark Lee Email & Phone Number

New York, United States

- Developed tool for latency benchmarking, PyTorch profile tracing, and CPU / GPU bottleneck detecting- Optimized production-deployed model for total inference speedup of 1.6x- Reduced floating point precision to halve memory footprint and deploy 2x as many models

- Developed filters to remove false obstacles from depth images, solving 5% of robot stoppage causes- Applied convolutional filters, morphological operations, and breadth-first search with priority queue- Optimized C++ code by 6x by reimplementing OpenCV logic with more concurrent operations

Greater San Diego Area

- Trained classifiers in PyTorch on > 1 million images and sensor metadata of zooplankton with 95+% test accuracy- Engineered system of training models and performing queried statistical analyses

Yeonsu-Gu, Incheon, Korea

- Extended ONNX (model converter) through bidirectional interface to Darknet (neural network C library)- Wrote C to Python interface using ctypes to map Darknet’s network to ONNX’s generic model representation

Bachelor Of Science (B.S.), Computer Science, Cumulative Gpa: 3.983 (Summa Cum Laude)

Activities and Societies: Jacobs School Scholar, Regents ScholarNotable courses taken: CSE 250A: Principles of AI: Probabilistic Reasoning.

3.96