Image Similarity in PowerPoint Designer

Microsoft Computer Science, 2021–22

Liaison(s): Michael Bentley, Jon Ko, Sarah Stein ’12
Advisor(s): Calden Wloka
Students(s): Devika Chipalkatti (PM-F), Emily Lu (PM-S), Ben Lucero, Sofiane Dissem, Swamik Lamichhane, Kendah Abughararh

Microsoft’s PowerPoint Designer suggests slide templates and images based on user input and historical customer data. Occasionally Microsoft purchases new images, but these images are difficult to recommend because they lack user data. Our goal is to aid Microsoft in diversifying image recommendations by providing a method for ranking image similarity between new images and images that have user data. The team explores machine learning and clustering techniques in an effort to recognize both concrete and abstract features that can be shared between images.