Designing Perceptual Puzzles by Differentiating Probabilistic Programs
Event Type
Technical Paper
Interest Areas
Research & Education
Presentation Types
Registration Categories
Full Conference Supporter
Full Conference
Virtual Conference Supporter
Virtual Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
DescriptionWe design new examples of visual illusions like "The Dress" by finding "adversarial examples" for principled models of human perception — specifically probabilistic models, which treat vision as Bayesian inference. To perform this search efficiently, we design a differentiable probabilistic programming language, exposing MCMC inference as a first-class differentiable function.