# Group equivariant neural posterior estimation

**Authors:** Maximilian Dax, Stephen R. Green, Jonathan Gair, Michael Deistler, Bernhard Schölkopf, Jakob H. Macke  
**arXiv:** [2111.13139](https://arxiv.org/abs/2111.13139)  
**Date:** 2021-11-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2111.13139v2/

## Key points

- Authors introduce a novel method for incorporating equivariances in simulation-based inference using conditional neural density estimators
- The proposed method, group equivariant neural posterior estimation (GNPE), integrates equivariances into expressive inference networks through joint transformations of parameters and data
- GNPE focuses on standardizing the "pose" of the data while estimating the posterior over parameters
- GNPE is architecture-independent and can accommodate both exact and approximate equivariances
- Authors apply GNPE to amortized inference of astrophysical binary black hole systems from gravitational-wave observations
- GNPE achieves state-of-the-art accuracy in inference tasks and significantly reduces inference times by three orders of magnitude compared to existing methods

## Summary

In their paper titled "Group equivariant neural posterior estimation," authors Maximilian Dax, Stephen R. Green, Jonathan Gair, Michael Deistler, Bernhard Schölkopf, and Jakob H. Macke introduce a novel method for incorporating equivariances in simulation-based inference using conditional neural density estimators. This new approach aims to address the limitations of existing methods by integrating equivariances into expressive inference networks through joint transformations of parameters and data. The proposed method, known as group equivariant neural posterior estimation (GNPE), focuses on standardizing the "pose" of the data while simultaneously estimating the posterior over parameters. GNPE is architecture-independent and can accommodate both exact and approximate equivariances. To demonstrate the effectiveness of their approach in a real-world scenario, the authors apply GNPE to amortized inference of astrophysical binary black hole systems from gravitational-wave observations. Through their experiments, Dax et al. show that GNPE achieves state-of-the-art accuracy in inference tasks while significantly reducing inference times by three orders of magnitude compared to existing methods. This groundbreaking research not only advances the field of simulation-based inference but also highlights the importance of incorporating geometric properties such as equivariances in neural network models for more efficient and accurate scientific problem-solving.