Teaching Machine Olfaction in an Undergraduate Deep Learning Course: A Pilot Integrating Chemistry, Machine Learning, and Sensory Evaluation

We describe a pilot implementation of a classroom activity that introduces machine olfaction, a cutting-edge research topic, into an undergraduate deep learning course, STATS 315, at the University of Michigan. Previously, the course focused on text and image data, but we introduce smell as a new modality to expand students’ learning experiences. By incorporating hands-on activities using carefully selected odorants, we engaged students in the data collection process and highlighted the complexities and variability inherent in sensory data. We also exposed students to graph neural networks and the curated GoodScents-Leffingwell dataset. The module adopted a multidisciplinary approach that combines chemistry, machine learning, and sensory evaluation. This pilot provides initial evidence for the feasibility of incorporating machine olfaction into deep learning instruction in the educational context where we implemented the activity, and it suggests directions for future curricular development. We discuss practical challenges encountered and propose directions for improving and extending this educational activity.