Student Experience

Lanie Kropp wanted to use transfer learning to classify medical images for disease diagnosis. Her research focused on the trade-offs between dataset size and similarity, contributing to her deepening expertise in neural networks and research methodologies.
 

• Objective: Compare the effectiveness of transfer learning using large generic datasets, such as ImageNet, versus smaller, domain-specific datasets, like x-rays, for medical imaging tasks.
   
• Findings: Lanie discovered that larger, more generic datasets like ImageNet outperformed smaller ones, providing insights into the balance between data quantity and relevance.
   
• Reflection: “I gained a lot from this experience, especially in understanding transfer learning and neural networks. I’ve delved deeper into these topics than anything else in my master’s program, thanks to the extensive time and research I’ve dedicated to them.”

Sarah McDougall focused her project on leveraging machine learning to better identify risk factors for pregnancy-related complications. Through her work, she developed skills in handling large datasets and creating multi-label classifiers while navigating challenges such as mapping medical code systems and the complexities of data preprocessing.
 

• Objective: Use machine learning to identify risk factors for complications, like preterm delivery and hemorrhaging, based on clinical and social data.
   
• Findings: Sarah discovered that the models she developed tended to produce more false positives than false negatives. This suggests that patients should be vigilant about various risk factors and outcomes and not overlook any serious issues that could impact their health.
   
• Reflection: “Another value of the capstone is its potential to make a real-world impact. Depending on how far you progress in the project, you could submit your findings for publication or network with others in the same research area. The journey doesn’t have to end after the second semester.”

Steve Hong created Food Tusker, an innovative smart food management system designed to minimize household food waste through smart Tupperware that tracks expiration dates. With his project that combines hardware and software development, Steve plans to launch a Kickstarter campaign to bring this eco-friendly solution to market.
 

• Objective: Develop a hardware-software solution to reduce household food waste using smart Tupperware with expiration tracking.
   
• Findings: Steve developed a working prototype that includes hardware integration, a custom display, and a Flutter-based app for cross-platform compatibility. Steve plans to develop the project further and launch a Kickstarter campaign.
   
• Reflection: “It’s been a really interesting journey. The complexity taught me how to scope projects effectively and focus on incremental progress. It’s been fascinating—and I get excited to get up and work on it.”