Learning with Marginalized Corrupted Features
Explore machine learning techniques for handling finite data through a lecture on learning with marginalized corrupted features. Delve into this alternative approach to regularization and priors, which involves corrupting existing data to generate infinite training samples. Discover how this computationally tractable method leads to fast, generalizable algorithms that scale well to large datasets. Examine applications in risk minimization regularization and marginalized deep learning for document representations. Review experimental results in part of speech tagging, document classification, and image classification. Learn from Kilian Q. Weinberger, an award-winning Assistant Professor from Washington University in St. Louis, known for his research in high-dimensional data analysis, metric learning, and machine-learned web-search ranking.