- Research Program Mentor
PhD Doctor of Philosophy candidate
Computer science, artificial intelligence, computer vision, medical image analysis, reinforcement learning, simulation, scientific visualization, Python package development
Identifying Bias in User-facing Machine Learning Models
Machine learning models are ubiquitous in our digital lives, but their presence is not alway obvious. Recently, Twitter users discovered that the automatic cropping used to display preview images was biased toward white, male faces. This is an example of statistical bias arising out of our human bias; because the facial detection algorithm was trained on a dataset of predominantly white, male faces, it predicts similar faces with high confidence. This project challenges students to identify similar instances of bias in interactive digital media resulting from machine learning and AI. Students will perform controlled experiments to understand the extent of these biases and explore their consequences for end users. Potential deliverables range from a blog series to a research paper.
2D/3D Registration of Lego Bricks
Registration is the process of aligning two different measurements of the same object. For example, in computer vision, it is often desirable to stitch two photos together to create a panorama, thereby registering one photo with the overlapping portion of the other. 2D/3D registration is when an existing 3D model, like a CAD surface, is aligned with an image of the same object, either simulated or in the real world. This has applications in medical imaging, such as aligning a 2D X-ray image with a 3D CT scan. In this projects, students will curate a dataset for 2D/3D registration of LEGO bricks and train a keypoint detection algorithm to perform the registration. LEGOs are well suited to this task because 3D models and simulated images are readily available. Minimum deliverables include a dataset hosted on Kaggle and an arXiv paper describing it. The maximum deliverables includes a trained registration algorithm and accompanying paper.