On Averaging Rotations

In this paper two common approaches to averaging rotations are compared to a more advanced approach based on a Riemannian metric. Very often the barycenter of the quaternions or matrices that represent the rotations are used as an estimate of the mean. These methods neglect that rotations belong to a non-linear manifold and re-normalization or orthogonalization must be applied to obtain proper rotations. These latter steps have been viewed as ad hoc corrections for the errors introduced by assuming a vector space. The article shows that the two approximative methods can be derived from natural approximations to the Riemannian metric, and that the subsequent corrections are inherent in the least squares estimation.     

Elastic geodesic paths in shape space of parameterized surfaces

This paper presents a novel Riemannian framework for shape analysis of parameterized surfaces. In particular, it provides efficient algorithms for computing geodesic paths which, in turn, are important for comparing, matching, and deforming surfaces. The novelty of this framework is that geodesics are invariant to the parameterizations of surfaces and other shape-preserving transformations of surfaces. The basic idea is to formulate a space of embedded surfaces (surfaces seen as embeddings of a unit sphere in IR3) and impose a Riemannian metric on it in such a way that the reparameterization group acts on this space by isometries. Under this framework, we solve two optimization problems. One, given any two surfaces at arbitrary rotations and parameterizations, we use a path-straightening approach to find a geodesic path between them under the chosen metric. Second, by modifying a technique presented in [25], we solve for the optimal rotation and parameterization (registration) between surfaces. Their combined solution provides an efficient mechanism for computing geodesic paths in shape spaces of parameterized surfaces. We illustrate these ideas using examples from shape analysis of anatomical structures and other general surfaces.     

A Riemannian Framework for Tensor Computing

Tensors are nowadays a common source of geometric information. In this paper, we propose to endow the tensor space with an affine-invariant Riemannian metric. We demonstrate that it leads to strong theoretical properties: the cone of positive definite symmetric matrices is replaced by a regular and complete manifold without boundaries (null eigenvalues are at the infinity), the geodesic between two tensors and the mean of a set of tensors are uniquely defined, etc. We have previously shown that the Riemannian metric provides a powerful framework for generalizing statistics to manifolds. In this paper, we show that it is also possible to generalize to tensor fields many important geometric data processing algorithms such as interpolation, filtering, diffusion and restoration of missing data. For instance, most interpolation and Gaussian filtering schemes can be tackled efficiently through a weighted mean computation. Linear and anisotropic diffusion schemes can be adapted to our Riemannian framework, through partial differential evolution equations, provided that the metric of the tensor space is taken into account. For that purpose, we provide intrinsic numerical schemes to compute the gradient and Laplace-Beltrami operators. Finally, to enforce the fidelity to the data (either sparsely distributed tensors or complete tensors fields) we propose least-squares criteria based on our invariant Riemannian distance which are particularly simple and efficient to solve.     

Discrete heat kernel determines discrete Riemannian metric

The Laplace–Beltrami operator of a smooth Riemannian manifold is determined by the Riemannian metric. Conversely, the heat kernel constructed from the eigenvalues and eigenfunctions of the Laplace–Beltrami operator determines the Riemannian metric. This work proves the analogy on Euclidean polyhedral surfaces (triangle meshes), that the discrete heat kernel and the discrete Riemannian metric (unique up to a scaling) are mutually determined by each other. Given a Euclidean polyhedral surface, its Riemannian metric is represented as edge lengths, satisfying triangle inequalities on all faces. The Laplace–Beltrami operator is formulated using the cotangent formula, where the edge weight is defined as the sum of the cotangent of angles against the edge. We prove that the edge lengths can be determined by the edge weights unique up to a scaling using the variational approach.The constructive proof leads to a computational algorithm that finds the unique metric on a triangle mesh from a discrete Laplace–Beltrami operator matrix.     

An intrinsic observer for a class of Lagrangian systems

We propose a new design method of asymptotic observers for a class of nonlinear mechanical systems: Lagrangian systems with configuration (position) measurements. Our main contribution is to introduce a state (position and velocity) observer that is invariant under any changes of the configuration coordinates. The observer dynamics equations, as the Euler-Lagrange equations, are intrinsic. The design method uses the Riemannian structure defined by the kinetic energy on the configuration manifold. The local convergence is proved by showing that the Jacobian of the observer dynamics is negative definite (contraction) for a particular metric defined on the state-space, a metric derived from the kinetic energy and the observer gains. From a practical point of view, such intrinsic observers can be approximated, when the estimated configuration is close to the true one, by an explicit set of differential equations involving the Riemannian curvature tensor. These equations can be automatically generated via symbolic differentiations of the metric and potential up to order two. Numerical simulations for the ball and beam system, an example where the scalar curvature is always negative, show the effectiveness of such approximation when the measured positions are noisy or include high frequency neglected dynamics.     

混杂数据的多核几何平均度量学习

在机器学习和模式识别任务中,选择一种合适的距离度量方法是至关重要的.度量学习主要利用判别性信息学习一个马氏距离或相似性度量.然而,大多数现有的度量学习方法都是针对数值型数据的,对于一些有结构的数据（比如符号型数据）,用传统的距离度量来度量两个对象之间的相似性是不合理的;其次,大多数度量学习方法会受到维度的困扰,高维度使得训练时间长,模型的可扩展性差.提出了一种基于几何平均的混杂数据度量学习方法.采用不同的核函数将数值型数据和符号型数据分别映射到可再生核希尔伯特空间,从而避免了特征的高维度带来的负面影响.同时,提出了一个基于几何平均的多核度量学习模型,将混杂数据的度量学习问题转化为求黎曼流形上两个点的中心点问题.在UCI数据集上的实验结果表明,针对混杂数据的多核度量学习方法与现有的度量学习方法相比,在准确性方面展现出更优异的性能.     

Restricted rotation distance between binary trees

Restricted rotation distance between pairs of rooted binary trees measures differences in tree shape and is related to rotation distance. In restricted rotation distance, the rotations used to transform the trees are allowed to be only of two types. Restricted rotation distance is larger than rotation distance, since there are only two permissible locations to rotate, but is much easier to compute and estimate. We obtain linear upper and lower bounds for restricted rotation distance in terms of the number of interior nodes in the trees. Further, we describe a linear-time algorithm for estimating the restricted rotation distance between two trees and for finding a sequence of rotations which realizes that estimate. The methods use the metric properties of the abstract group known as Thompson's group F.     

Regularizing Flows over Lie Groups

In this paper we discuss regularization of images that take their value in matrix Lie groups. We describe an image as a section in a principal bundle which is a fibre bundle where the fiber (the feature space) is a Lie group. Via the scalar product on the Lie algebra, we define a bi-invariant metric on the Lie-group manifold. Thus, the fiber becomes a Riemannian manifold with respect to this metric. The induced metric from the principal bundle to the image manifold is obtained by means of the bi-invariant metric. A functional over the space of sections, i.e., the image manifolds, is defined. The resulting equations of motion generate a flow which evolves the sections in the spatial-Lie-group manifold. We suggest two different approaches to treat this functional and the corresponding PDEs. In the first approach we derive a set of coupled PDEs for the local coordinates of the Lie-group manifold. In the second approach a coordinate-free framework is proposed where the PDE is defined directly with respect to the Lie-group elements. This is a parameterization-free method. The differences between these two methods are discussed. We exemplify this framework on the well-known orientation diffusion problem, namely, the unit-circle S ¹ which is identified with the group of rotations in two dimensions, SO(2). Regularization of the group of rotations in 3D and 4D, SO(3) and SO(4), respectively, is demonstrated as well.

Geometry of quantum state space and quantum correlations

Quantum state space is endowed with a metric structure, and Riemannian monotone metric is an important geometric entity defined on such a metric space. Riemannian monotone metrics are very useful for information-theoretic and statistical considerations on the quantum state space. In this article, considering the quantum state space being spanned by \(2\times 2\) density matrices, we determine a particular Riemannian metric for a state \(\rho \) and show that if \(\rho \) gets entangled with another quantum state, the negativity of the generated entangled state is, upto a constant factor, equal to square root of that particular Riemannian metric . Our result clearly relates a geometric quantity to a measure of entanglement. Moreover, the result establishes the possibility of understanding quantum correlations through geometric approach.     

Metric learning for text documents

Many algorithms in machine learning rely on being given a good distance metric over the input space. Rather than using a default metric such as the Euclidean metric, it is desirable to obtain a metric based on the provided data. We consider the problem of learning a Riemannian metric associated with a given differentiable manifold and a set of points. Our approach to the problem involves choosing a metric from a parametric family that is based on maximizing the inverse volume of a given data set of points. From a statistical perspective, it is related to maximum likelihood under a model that assigns probabilities inversely proportional to the Riemannian volume element. We discuss in detail learning a metric on the multinomial simplex where the metric candidates are pull-back metrics of the Fisher information under a Lie group of transformations. When applied to text document classification the resulting geodesic distance resemble, but outperform, the tfidf cosine similarity measure.     

Three-dimensional anisotropic geometric metrics based on local domain curvature and thickness

A three-dimensional anisotropic Riemannian metric is constructed from a triangulated CAD model to control its spatial discretization for numerical analysis. In addition to the usual curvature criterion, the present geometric metric is also based on the local thickness of the modeled domain. This local thickness is extracted from the domain skeleton while local curvature is deduced from the model triangulated boundaries. A Cartesian background octree is used as the support medium for this metric and skeletonization takes advantage of this structure through an octree extension of a digital medial axis transform. For this purpose, the octree has to be refined according to not only boundary curvature but also a local separation criterion from digital topology theory. The resulting metric can be used to geometrically adapt any mesh type as long as metric-based adaptation tools are available. To illustrate such an application, geometric adaptation of overlay meshes used in grid-based methods for unstructured hexahedral mesh generation is presented. However, beyond mesh generation, the present metric may also be useful as a shape analysis tool and such a possibility could be explored in future developments.     

Intrinsic Polynomials for Regression on Riemannian Manifolds

We develop a framework for polynomial regression on Riemannian manifolds. Unlike recently developed spline models on Riemannian manifolds, Riemannian polynomials offer the ability to model parametric polynomials of all integer orders, odd and even. An intrinsic adjoint method is employed to compute variations of the matching functional, and polynomial regression is accomplished using a gradient-based optimization scheme. We apply our polynomial regression framework in the context of shape analysis in Kendall shape space as well as in diffeomorphic landmark space. Our algorithm is shown to be particularly convenient in Riemannian manifolds with additional symmetry, such as Lie groups and homogeneous spaces with right or left invariant metrics. As a particularly important example, we also apply polynomial regression to time-series imaging data using a right invariant Sobolev metric on the diffeomorphism group. The results show that Riemannian polynomials provide a practical model for parametric curve regression, while offering increased flexibility over geodesics.     

Image Manifolds which are Isometric to Euclidean Space

Recently, the Isomap procedure [10] was proposed as a new way to recover a low-dimensional parametrization of data lying on a low-dimensional submanifold in high-dimensional space. The method assumes that the submanifold, viewed as a Riemannian submanifold of the ambient high-dimensional space, is isometric to a convex subset of Euclidean space. This naturally raises the question: what datasets can reasonably be modeled by this condition? In this paper, we consider a special kind of image data: families of images generated by articulation of one or several objects in a scene—for example, images of a black disk on a white background with center placed at a range of locations. The collection of all images in such an articulation family, as the parameters of the articulation vary, makes up an articulation manifold, a submanifold of L ². We study the properties of such articulation manifolds, in particular, their lack of differentiability when the images have edges. Under these conditions, we show that there exists a natural renormalization of geodesic distance which yields a well-defined metric. We exhibit a list of articulation models where the corresponding manifold equipped with this new metric is indeed isometric to a convex subset of Euclidean space. Examples include translations of a symmetric object, rotations of a closed set, articulations of a horizon, and expressions of a cartoon face. The theoretical predictions from our study are borne out by empirical experiments with published Isomap code. We also note that in the case where several components of the image articulate independently, isometry may fail; for example, with several disks in an image avoiding contact, the underlying Riemannian manifold is locally isometric to an open, connected, but not convex subset of Euclidean space. Such a situation matches the assumptions of our recently-proposed Hessian Eigenmaps procedure, but not the original Isomap procedure.     

基于对数欧氏度量学习的概率黎曼空间量化方法

在许多机器学习应用中,需要分析的数据可能由对称正定矩阵构成,而经典的欧氏机器学习算法处理这种数据的性能较差。针对此问题,提出一种新的基于对数欧氏度量学习的概率黎曼空间量化方法。该方法将对称正定矩阵看做对数欧氏度量下黎曼流形上的点,采用对数欧氏度量学习距离函数将概率学习矢量量化方法从欧氏空间推广到对称正定黎曼空间。在BCI IV 2a脑电数据集上,该方法相较于概率学习矢量量化方法识别正确率提升20%,高于竞赛第一名;并且计算速度快,模型训练及测试时间分别为基于仿射不变度量的同类型算法的1%和10%。在BCI III IIIa和图像数据集ETH-80上也取得了较好的结果。     

Application of the Fisher-Rao Metric to Structure Detection

Certain structure detection problems can be solved by sampling a parameter space for the different structures at a finite number of points and checking each point to see if the corresponding structure has a sufficient number of inlying measurements. The measurement space is a Riemannian manifold and the measurements relevant to a given structure are near to or on a submanifold which constitutes the structure. The probability density function for the errors in the measurements is described using a generalisation of the Gaussian density to Riemannian manifolds. The conditional probability density function for the measurements yields the Fisher information which defines a metric, known as the Fisher-Rao metric, on the parameter space. The main result is a derivation of an asymptotic approximation to the Fisher-Rao metric, under the assumption that the measurement noise is small. Using this approximation to the Fisher-Rao metric, the parameter space is sampled, such that each point of the parameter space is near to at least one sample point, to within the level of accuracy allowed by the measurement errors. The probability of a false detection of a structure is estimated. The feasibility of this approach to structure detection is tested experimentally using the example of line detection in digital images.     

Essential Matrix Estimation Using Gauss-Newton Iterations on a Manifold

A novel approach for essential matrix estimation is presented, this being a key task in stereo vision processing. We estimate the essential matrix from point correspondences between a stereo image pair, assuming that the internal camera parameters are known. The set of essential matrices forms a smooth manifold, and a suitable cost function can be defined on this manifold such that its minimum is the desired essential matrix. We seek a computationally efficient optimization scheme towards meeting the demands of on-line processing of video images. Our work extends and improves the earlier research by Ma et al., who proposed an intrinsic Riemannian Newton method for essential matrix computations. In contrast to Ma et al., we propose three Gauss-Newton type algorithms that have improved convergence properties and reduced computational cost. The first one is based on a novel intrinsic Newton method, using the normal Riemannian metric on the manifold consisting of all essential matrices. The other two methods are Newton-like methods, that are more efficient from a numerical point of view. Local quadratic convergence of the algorithms is shown, based on a careful analysis of the underlying geometry of the problem.     

Evaluating methods to simulate crop rotations for climate impact assessments – A case study on the Canterbury plains of New Zealand

Most climate impact assessments for food production simulate single crops with re-initialised soil conditions. However, crop rotations with multiple crops are used in many agricultural regions worldwide. This case-study compares methods to aggregate outputs from simulations of multi-crop systems for climate impact assessments. The APSIM model was used to simulate four crops as monocultures (re-initialised or continuous) or as (single or multiple) instances of continuous rotations. We considered two contrasting climates and two soil types, with four production intensification scenarios (high/low water and nitrogen input). Results suggest that differences among the methods depend on the impact variable of interest and the degree of intensification. Detailed simulations (i.e. multiple runs of continuous rotations) were especially valuable for soil-related variables and limiting growth conditions. These results can indicate sources of uncertainty for large scale impact and adaptation assessments where simplifications of crop rotations are often necessary.     

黎曼流形上的保局投影在图像集匹配中的应用

目的提出了黎曼流形上局部结构特征保持的图像集匹配方法。方法该方法使用协方差矩阵建模图像集合,利用对称正定的非奇异协方差矩阵构成黎曼流形上的子空间,将图像集的匹配转化为流形上的点的匹配问题。通过基于协方差矩阵度量学习的核函数将黎曼流形上的协方差矩阵映射到欧几里德空间。不同于其他方法黎曼流形上的鉴别分析方法,考虑到样本分布的局部几何结构,引入了黎曼流形上局部保持的图像集鉴别分析方法,保持样本分布的局部邻域结构的同时提升样本的可分性。结果在基于图像集合的对象识别任务上测试了本文算法,在ETH80和YouTube Celebrities数据库分别进行了对象识别和人脸识别实验,分别达到91.5%和65.31%的识别率。结论实验结果表明,该方法取得了优于其他图像集匹配算法的效果。     

Uncertainty evaluation and model selection of extreme learning machine based on Riemannian metric

Considering the uncertainty of hidden neurons, choosing significant hidden nodes, called as model selection, has played an important role in the applications of extreme learning machines(ELMs). How to define and measure this uncertainty is a key issue of model selection for ELM. From the information geometry point of view, this paper presents a new model selection method of ELM for regression problems based on Riemannian metric. First, this paper proves theoretically that the uncertainty can be characterized by a form of Riemannian metric. As a result, a new uncertainty evaluation of ELM is proposed through averaging the Riemannian metric of all hidden neurons. Finally, the hidden nodes are added to the network one by one, and at each step, a multi-objective optimization algorithm is used to select optimal input weights by minimizing this uncertainty evaluation and the norm of output weight simultaneously in order to obtain better generalization performance. Experiments on five UCI regression data sets and cylindrical shell vibration data set are conducted, demonstrating that the proposed method can generally obtain lower generalization error than the original ELM, evolutionary ELM, ELM with model selection, and multi-dimensional support vector machine. Moreover, the proposed algorithm generally needs less hidden neurons and computational time than the traditional approaches, which is very favorable in engineering applications.