Calvin Murdock
PhD Student
Machine Learning
Monday, December 7
3:00 pm to 4:00 pm
3:00 pm to 4:00 pm
Semantic Component Analysis
Event Location: NSH 1507
Abstract: We propose a novel formulation for component analysis that allows for rich instance-level constraints that encourage semantic interpretability of the learned components. Even with simple features and intuitive spatial consistency priors, our method produces accurate, semantically-meaningful image segmentations both with and without supervision.