The goal of this SuperUROP is to develop machine learning models and datasets capable of recognizing and interpreting mixtures of smells. Existing approaches focus on identifying single substances, but real-world odors are complex combinations of volatile compounds that interact non-linearly. We will collect sensor data from controlled mixtures across categories such as fruits, herbs, spices, and household chemicals to create a benchmark dataset for mixture reasoning.

Using this data, we will train attention-based and multi-label models to classify mixtures, estimate composition, and generalize to novel combinations. Building on our previous work SmellNet, this project aims to advance mixture-aware AI for smell, with all datasets and models released as open-source resources to support the broader olfactory machine learning community.

I chose to pursue SuperUROP to gain structured research experience while working on groundbreaking projects. My previous UROP experiences prepared me to undertake a SuperUROP. This year, I hope to learn more about multimodal learning, the process of publishing, and collaboration. I am most excited for how much I will grow from this experience and for opportunities to publish.