Scalable Reversible Generative Models with Free-form Continuous Dynamics
By Will Grathwohl. 2018
Subfields: Generative Models
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Generative Models
- Balaji Lakshminarayanan
- Yee Whye Teh
- David Kristjanson Duvenaud
- Stefano Ermon
- Jörn-Henrik Jacobsen
- Eric T. Nalisnick
- Richard S. Zemel
- Max Welling
Citations in the corpus
- Excessive Invariance Causes Adversarial Vulnerability (2018, background): “…that ResNets and iRevNets behave similarly on various levels of their representation on challenging tasks (Jacobsen et al., 2018) and that iRevNets as well as Glow-type networks are related to ResNets by the choice of dimension splitting applied in their residual blocks (Grathwohl et al., 2019).”
- Invertible Residual Networks (2018, background): “, 2017), Glow (Kingma & Dhariwal, 2018) and FFJORD (Grathwohl et al., 2019).”
- Building Normalizing Flows with Stochastic Interpolants (2022, unknown): “(2019), Free-form continuous flows (FFJORD) (Grathwohl et al., 2019), and OT-Flow (Finlay et al.”
- Riemannian Score-Based Generative Modeling (2022, methodology): “They extend continuous normalizing flows (CNFs) (Grathwohl et al., 2019) to the manifold setting, by implicity parametrizing flows as solutions of Ordinary Differential Equations (ODEs).”
- Hybrid Models with Deep and Invertible Features (2019, background): “While this term could be expensive to compute for general functions, much recent work has been done on defining expressive invertible neural networks with easy-to-evaluate Jacobians (Dinh et al., 2015; 2017; Kingma & Dhariwal, 2018; Grathwohl et al., 2019).”
- Diffusion Models: A Comprehensive Survey of Methods and Applications (2022, background): “The theory of neural ODEs [30] and continuous normalizing flows [77] indicates that pode θ can be computed accurately albeit with high computational cost.”
- Stochastic Interpolants: A Unifying Framework for Flows and Diffusions (2023, methodology): “At the heart of progress is the idea to apply ordinary or stochastic differential equations (ODE/SDE) to continuously transform samples from a base probability density function (PDF) ρ0 into samples from a target density ρ1 (or vice-versa), and the realization that inference over the velocity field ”
- IDF++: Analyzing and Improving Integer Discrete Flows for Lossless Compression (2020, background): “Examples include autoregressive neural networks [36, 42], variational auto-encoders [26, 41, 45, 31] and normalizing flows [47, 46, 10, 41, 11, 37, 27, 21, 15, 14, 20, 17], with the latter two enjoying faster decoding times.”