This project aims at developing some few-shot learning techniques for adapting a deep neural network model pretrained on some existing scenes for a video-based application to a new scene using very few labeled examples from that scene. Due to the complexity of the project, it is expected to take more than one academic term to complete.
UROP1100 UROP2100 UROP3100 UROP4100
Students working on this project require strong background in both quantitative reasoning (mathematics) and software development (computer science). Prior experience in machine learning and in software development on Linux platforms would be an advantage.
Applicant's Learning Objectives:
The main learning objectives include formulating research problems, identifying and applying relevant machine learning and AI techniques for solving the problems, designing and conducting experiments, and reporting results.