M22.5 CONNECT THE MECHANISM
Represent probability through invertible maps or energy scores
Stretch a bell curve and its height must drop. Score states instead and you face a sum too big to compute. Two exact-probability ideas, and the problem each one dodges.
LESSON OVERVIEW15 min lesson
Lesson overview
Stretch a bell curve and its height must drop. Score states instead and you face a sum too big to compute. Two exact-probability ideas, and the problem each one dodges.
What you’ll explore
- Normalizing flows transform a known density with invertible maps and volume corrections; energy-based models assign unnormalized scores and require a way to handle normalization and sampling.
GO TO THE SOURCE
Original explanations, connected to the research.
NICE: Non-linear Independent Components Estimation (Dinh, Krueger & Bengio, 2014)Density estimation using Real NVP (Dinh, Sohl-Dickstein & Bengio, 2016)Glow: Generative Flow with Invertible 1x1 Convolutions (Kingma & Dhariwal, 2018)A Tutorial on Energy-Based Learning (LeCun et al., 2006)Deep Learning, chapter 18: Confronting the Partition Function (Goodfellow, Bengio & Courville, 2016)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.