This repository is a curated list of resources for understanding and applying normalizing flows (NF), a powerful statistical tool for constructing complex probability distributions. It serves researchers, engineers, and practitioners in machine learning and statistics interested in generative modeling, density estimation, and Bayesian inference. The primary benefit is a comprehensive, up-to-date overview of the field, including foundational papers, recent advancements, and practical implementations.

How It Works

This is a curated list, not a software library. It organizes resources into categories such as publications, applications, videos, packages, and code repositories. The content is structured to provide a historical overview and highlight key developments, such as the introduction of coupling layers (NICE, RealNVP), autoregressive flows (MAF, IAF), continuous-time flows (FFJORD), and equivariant flows. The list emphasizes advancements in architectural components like invertible convolutions and spline-based transformations.

Quick Start & Requirements

This repository does not require installation or execution. It is a collection of links and information.

Highlighted Details

• Extensive coverage of over 60 publications, detailing advancements from 2014 to the present.
• Categorized lists of 15+ software packages across PyTorch, TensorFlow, and JAX, and 18+ code repositories.
• Includes 8+ video tutorials and lectures from leading researchers in the field.
• Features applications in diverse areas like time series analysis, molecular modeling, and reinforcement learning.

Maintenance & Community

The repository is community-driven, with a clear contribution process outlined for adding new resources. It encourages community input via issues and discussions. The README is auto-generated from YAML data files.

Licensing & Compatibility

The repository itself is licensed under the MIT License, allowing for broad use and modification. The linked resources may have their own licenses.

Limitations & Caveats

As a curated list, it does not provide direct software functionality. The quality and maintenance of linked packages and repositories vary, requiring users to evaluate them individually. The field is rapidly evolving, so the list may not capture the absolute latest developments immediately.