Explore the fascinating world of machine learning applied to scent in this 39-minute lecture from MIT's Introduction to Deep Learning course. Delve into the process of digitizing smell, understand the complexities of the human olfactory system, and learn how to set up the problem of predicting odor descriptors. Discover the molecule fragrance dataset and various baseline algorithms before diving into graph neural networks and their application to molecular structures. Investigate the odor embedding space, molecular neighbors, and generalization techniques. Gain insights into explaining and interpreting predictions in this cutting-edge field. Conclude with a summary of current progress and potential future developments in machine learning for scent analysis.
Overview
Syllabus
- Introduction
- Digitizing smell
- The sense of smell
- Problem setup
- Molecule fragrance dataset
- Baseline algorithms
- Graph neural networks
- Molecules to graphs
- Predicting odor descriptors
- The odor embedding space
- Molecular neighbors
- Generalization
- Explaining/interpreting predictions
- Summary and future work
Taught by
https://www.youtube.com/@AAmini/videos
