Flow Models: Normalizing Flows | AI Knowledge Hub |…

Normalizing Flows for Generation

Flow models (Normalizing Flows) are a class of generative models that learn an exact probability distribution through a sequence of invertible transformations. Unlike GANs and VAEs, flow models learn the explicit probability distribution, allowing for precise calculation of likelihood and data generation with full control.

This makes them particularly useful for applications where accurate probability estimation is important, such as density estimation, anomaly detection, and Bayesian inference.

Sequence of Invertible Transformations

Transforming a simple distribution (e.g., normal) into a complex one.

The change of variables formula is used to calculate the likelihood.

live demo · related simulation● LIVE

RealNVP: Coupling Layers

Glow: Invertible 1x1 convolutions

FFJORD: Continuous normalizing flows

Frequently asked questions

What are the advantages of flow models compared to GANs and VAEs?

Flow models learn an explicit probability distribution, enabling precise likelihood calculation, full control over data generation, and use in density estimation. They also exhibit more stable training compared to other generative methods.

What are the disadvantages of flow models?

Flow models can be computationally intensive due to the complex transformations involved, and their performance depends heavily on the quality of the learned transformation functions. Further research is ongoing to address these challenges.

▶ Try it live

Everything above runs in your browser — open Hash Function Avalanche Visualizer and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.

▶ Open Hash Function Avalanche Visualizer simulation