Daphne Tsai Email & Phone Number

Pittsburgh, Pennsylvania, United States

Perception for autonomous drones:)

Ann Arbor, Michigan, United States

• Overseeing course operations by making homework assignments for ~700 students and managing 16 instructional aides

Pittsburgh, Pennsylvania, United States

• Developed a custom ONNX configuration for a CUDA-accelerated, transformer-based semantic segmentation model and improved inference time by 44% through TensorRT optimization and PTQ static methods
• Wrote a custom MobileNetV2 backbone architecture to optimize real-time semantic segmentation on PyTorch 2.0 by 64% while under GPU computing constraints to investigate accuracy vs. speed tradeoff
• Fine-tuned a powerline detection model on a training set of 10122 simulation and real-world IR images that achieved a 0.312 IoU after validation on 1206 real-world IR images through 100 epochs

Ann Arbor, Michigan, United States

• Faculty Advisor: Professor Nikhil Bansal
• Conducted an extensive literature review on past and current algorithms used to approximate the k-set cover problem

Ann Arbor, Michigan, United States

• Hosted 3.5 hours of office hours per week, taught ~20 students weekly, and answered student questions on Piazza
• Collaborated with a 25-person teaching team to write, test, and grade exams for a class of ~700 students

Ann Arbor, Michigan, United States

• Trained a YOLOv4 model that achieved a mAP score of 48.79% (IoU threshold of 0.5) at 70.19 FPS on proprietary testing data to be integrated onto the company’s L4 autonomous driving software platform
• Fused the trained YOLOv4 model with existing V2X infrastructure (SPaT messages) into a ROS Melodic environment to improve decision making at traffic light intersections
• Conducted extensive literature reviews on deep learning and V2X technology integration for novel perception models

• Faculty Advisors: Dr. Carol Flannagan and Dr. Kathleen Klinich, CMISST group
• Developed a data evaluation program to parse accuracy scores of 7587 IVBSS videos to evaluate neural network strength
• Expanded upon labeling software by implementing a limb-labeling feature to track passenger movement
• Discussed research findings to faculty and students at the University’s Engineering Research Symposium

Toledo, Ohio, United States

• Leveraged Node.js and Puppeteer to scrape web data by refactoring API code for Lighthouse and Google Search Console
• Created data visualizations of SEO trends using PostgreSQL while working in an agile development environment
• Presented project findings and suggestions for future work to executive leadership members and other interns

Master Of Science In Engineering - Mse, Computer Science

Activities and Societies: V1 Michigan, M-Fly Aero Design, Graduate Society of Women Engineers, Eta Kappa NuRelevant Coursework: Robotic.

Bachelor Of Science In Engineering - Bse, Computer Science

Activities and Societies: M-fly Aero Design, V1 Michigan, Society of Women Engineers, Multidisciplinary Design Program, Perot Jain TechLab.