Parameter Estimation in Spherical Symmetry Groups.

Yu-Hui Chen, Dennis L. Wei, Gregory E. Newstadt, Marc De Graef, Jeff P. Simmons, Alfred O. Hero IiiIEEE Signal Processing Letters 2015 · OpenReview

TL;DR

This paper introduces a novel method for estimating parameters of probability distributions that exhibit symmetry under a finite group of transformations. They extend the Von Mises-Fisher (VMF) distribution to handle these symmetries, creating a "G-invariant VMF," and use an Expectation-Maximization (EM) algorithm for efficient parameter estimation. This approach significantly improves the accuracy of estimating crystal orientations from EBSD data, a crucial task in materials science.

What Problem Does This Paper Address?

Estimating the parameters of probability distributions is a fundamental task in many scientific fields. However, when the data's underlying distribution has symmetries, standard estimation methods can become inefficient or even inaccurate. This is particularly true in materials science, where the orientation of crystal grains within a material is often described by angles or quaternions, and these orientations are subject to the inherent symmetries of the crystal structure (e.g., cubic, hexagonal).

For instance, when analyzing crystal orientations using Electron Backscatter Diffraction (EBSD), the same physical orientation can be represented by multiple equivalent sets of angles due to the crystal's symmetry. This ambiguity complicates the estimation of the "true" mean orientation and how tightly the orientations are clustered around it. Existing methods either ignore these symmetries, leading to suboptimal results, or try to simplify the problem by mapping orientations to a "fundamental zone," which can discard valuable information. This paper aims to address this by developing a principled statistical framework that explicitly accounts for these group symmetries.

Methods

The core contribution of this paper is the development of a G-invariant extension of the Von Mises-Fisher (VMF) distribution. The VMF distribution is a standard model for directional data on a sphere. The authors first establish a general theorem showing that any probability distribution invariant under a finite topological group can be represented as a finite mixture of its "translated" versions, where the number of components in the mixture equals the order of the group, and all components share the same parameters.

When applied to spherical symmetry groups, this leads to the G-invariant VMF distribution. This distribution is parameterized by a mean orientation and a concentration parameter, but it accounts for the group's symmetries by effectively averaging over all equivalent representations of the mean orientation.

To estimate the parameters of this G-invariant VMF distribution, the authors employ the Expectation-Maximization (EM) algorithm. This is a standard iterative technique for finding maximum likelihood estimates in models with latent variables, which is well-suited for finite mixture models. The EM algorithm iteratively refines estimates of the mean orientation and concentration by alternating between an "E-step" (calculating expected values of latent variables) and an "M-step" (maximizing the expected log-likelihood).

Datasets

The paper evaluates its proposed method using:

  • Simulated data: They generated i.i.d. samples from the G-invariant VMF distribution with known parameters to compare the performance of their estimator against existing methods under controlled conditions.
  • Real EBSD data: They used data acquired from a polycrystalline Nickel alloy sample (IN100), which exhibits cubic symmetry (m3m point symmetry group). This real-world dataset allowed them to demonstrate the practical advantages of their approach in a materials science application.

The paper does not introduce a new benchmark dataset but rather applies its method to a relevant real-world dataset.

Key Insights and Ideas

  • Symmetry as a Mixture: The fundamental insight is that group-invariant probability distributions can be elegantly represented as finite mixtures of their group-translated counterparts. This provides a powerful theoretical bridge between group theory and statistical modeling.
  • G-invariant VMF: Extending the VMF distribution to incorporate group symmetries directly addresses the ambiguity in directional data arising from symmetries, leading to more accurate and robust parameter estimation.
  • EM for G-invariant Models: The EM algorithm is a natural and effective tool for estimating parameters of these G-invariant mixture distributions, offering an iterative approach that converges to maximum likelihood estimates.
  • Fundamental Zone Limitations: The paper highlights that simply mapping data to a fundamental zone can lead to information loss, and their proposed method, by using the full mixture representation, avoids this drawback.

Key Results

  • Improved Orientation Estimation: The proposed EM-ML estimator for the G-invariant VMF distribution achieves significantly more accurate estimation of mean crystal orientations compared to naive VMF estimators and a modified VMF estimator that uses fundamental zones.
  • Faster Convergence: The G-invariant EM-ML estimator recovers the true mean orientation much faster as the concentration parameter increases (i.e., as the data becomes less dispersed).
  • Reduced Bias in Concentration: The bias in estimating the concentration parameter is significantly lower for the proposed EM-ML estimator compared to other methods.
  • Practical Application: The method demonstrated superior performance on real EBSD data from a Nickel alloy, showing more homogeneous and accurate grain orientation estimates, even in regions where standard OEM software produced inconsistent results.

Limitations and Future Work

  • Computational Cost: While the EM algorithm is efficient, the computational cost can increase with the order of the symmetry group, as the number of mixture components grows.
  • Generalization to Other Groups: The paper focuses on spherical symmetry groups. The authors suggest that applying the finite mixture representation to other types of groups would be a worthwhile direction for future work.
  • Assumptions: The method assumes the underlying distribution is indeed G-invariant. Deviations from this assumption might affect performance.

The authors explicitly state that extending their framework to other types of groups is a promising avenue for future research.

Summary updated

Abstract

This letter considers statistical estimation problems where the probability distribution of the observed random variable is invariant with respect to actions of a finite topological group. It is shown that any such distribution must satisfy a restricted finite mixture representation. When specialized to the case of distributions over the sphere that are invariant to the actions of a finite spherical symmetry group G, a group-invariant extension of the Von Mises Fisher (VMF) distribution is obtained. The G-invariant VMF is parameterized by location and scale parameters that specify the distribution's mean orientation and its concentration about the mean, respectively. Using the restricted finite mixture representation these parameters can be estimated using an Expectation Maximization (EM) maximum likelihood (ML) estimation algorithm. This is illustrated for the problem of mean crystal orientation estimation under the spherically symmetric group associated with the crystal form, e.g., cubic or octahedral or hexahedral. Simulations and experiments establish the advantages of the extended VMF EM-ML estimator for data acquired by Electron Backscatter Diffraction (EBSD) microscopy of a polycrystalline Nickel alloy sample.
BibTeX
@article{chen2015parameter,
  author = {Yu-Hui Chen and Dennis L. Wei and Gregory E. Newstadt and Marc De Graef and Jeff P. Simmons and Alfred O. Hero III},
  title = {{Parameter Estimation in Spherical Symmetry Groups.}},
  journal = {IEEE Signal Processing Letters},
  year = {2015},
  url = {https://openreview.net/forum?id=-6T1Bkp1O37u},
}
Figure