CS Paper Radar

Closed-Form Gaussian Information Projection for Embedded Hypercubes and Beta--Gaussian Mixtures

Ali Darijani, Emadedin Soufian, Jürgen Beyerer

cs.IT ·submitted 2026-11-11 21:28 UTC · v1
theory

Abstract

We derive closed-form Gaussian information projections of Beta--Gaussian distributions whose tangential components are supported on affine embedded hypercubes and whose normal components are Gaussian. The optimal Gaussian is uniquely characterized by moment matching, and the corresponding minimum Kullback--Leibler divergence admits an explicit expression depending only on the intrinsic Beta parameters, independently of the embedding geometry, ambient dimension, and normal Gaussian parameters. In the uniform case, this divergence reduces to a linear function of the intrinsic dimension. We further extend this construction to finite, pure embedded cubical complexes by considering mixtures of the componentwise Gaussian projections. Using the joint convexity of the Kullback--Leibler divergence, we establish an explicit upper bound on the approximation error of the resulting Gaussian Mixture. For uniform tangential distributions and volume-based mixture weights, the bound is linear in the intrinsic dimension and independent of the geometry of the complex, its embedding, and the normal Gaussian parameters.

Comment
11 pages

Closest papers

Nearest neighbours by meaning across the whole index.

Closed-Form Gaussian Information Projection for Embedded Simplices and Simplicial Mixtures

cs.IT ·2026-11-09 ·Ali Darijani, Emadedin Soufian, Jürgen Beyerer
similarity 0.80
Abstract · 27 Pages, 3 Figures

We study Gaussian approximation of Dirichlet--Gaussian probability density functions on embedded simplices. We derive the unique information projection onto the family of nondegenerate ambient Gaussian probability density functions, obtaining closed-form expressions for the optimal Gaussian and the minimum divergence for arbitrary positive Dirichlet parameters. The minimum divergence depends only on the intrinsic simplex dimension and Dirichlet parameters, is invariant up to an affine isomorphism, independent of the ambient dimension, and independent of the normal covariance. Furthermore, by joint convexity of Kullback--Leibler divergence, we establish explicit intrinsic dimension-dependent sharp upper bounds for finite Gaussian Mixture with shared component weights. For uniform probability density functions, these bounds reduce to dimension-dependent constants. The results provide closed-form Gaussian Mixture constructions and rigorous approximation guarantees for probability density functions associated with embedded simplicial complexes. A numerical example on a circular curve illustrates the theoretical results and their connections to geometric Gaussian Mixture representations.

0.24

Closed-Form Gaussian Mixture Representations of d-Dimensional Triangulated Domains

cs.IT math.NAmath.PR ·2026-10-31 ·Ali Darijani, Benedikt Stratmann, Uwe D. Hanebeck, Ratan Bahadur Thapa, Jürgen Beyerer +1 more
publishedtheory similarity 0.71
Abstract · Accepted to "2026 IEEE Sensor Data Fusion: Trends, Solutions, Applications (SDF)". 17 pages and 6 figures

We derive a Gaussian Mixture (GM) representation of a full-dimensional subset of Euclidean space with finite volume from a finite simplicial triangulation in closed form. On each simplex, we replace the uniform probability density function (PDF) with a Gaussian PDF that minimizes the Kullback--Leibler (KL) divergence, arriving at the information projection onto the family of Gaussian PDFs. By exploiting affine isomorphisms between simplices, the joint convexity and bijection invariance properties of the KL divergence, identities for Dirichlet PDFs, and a partition-of-unity argument, we establish a dimension-dependent but geometry-independent sharp upper bound on the global KL divergence between the resulting GM PDF and the uniform PDF on the original domain. Numerical examples illustrate the construction and demonstrate its potential in practical applications.

0.59

Fisher Information and Support Complexity of Gaussian Mixtures

cs.IT ·2026-10-15 ·Luca Barletta, Alex Dytso
theory similarity 0.59
Abstract · 24 pages, single-column

We study lower bounds on the Fisher information of Gaussian location mixtures and their applications to estimation problems. First, for a discrete mixing distribution, we derive a universal positive-semidefinite lower bound on the Fisher information matrix in terms of the Rényi entropy of order two. For a mixing distribution supported on $K$ points, this bound scales as $\log(K)/K^2$, improving by a logarithmic factor on the support-size dependence obtained from Stam's inequality. We then develop dimension-aware trace bounds using classical Sobolev inequalities. These bounds are expressed in terms of Sibson's $α$-mutual information and yield the scaling $K^{-2/n}$ in dimension $n$. A complementary bound involving the $α$-mutual information of order $1/2$ is obtained using the sharp Young convolution inequality. The derived inequalities are further specialized to rotationally invariant mixing distributions supported on discrete spherical shells. In this setting, an intrinsic-dimensional bound on Sibson mutual information leads to lower bounds on the Fisher information in terms of the number of occupied shells. As applications, we study least favorable priors for bounded Gaussian mean estimation. In the scalar case, we show that the least favorable prior on $[-R,R]$ has at least $Ω(R\sqrt{\log R})$ support points and that each of its probability masses is of order at most $1/R$. In the vector case, we show that the number of occupied shells grows at least linearly with $R$. Finally, Brown's identity and the I-MMSE relationship are used to translate the Fisher-information inequalities into upper bounds on the MMSE and mutual information.

0.28

Beyond Unimodal Bases: Pullback Geometry for Multimodal Data

cs.LG ·2026-09-30 ·Honglei Brinkmann, Lucas Ng, Georgios Batzolis, Mark Girolami, Carola-Bibiane Schönlieb +1 more
similarity 0.56
Abstract

Data-driven Riemannian geometry provides nonlinear interpolation and geometric representations of high-dimensional data. For these operations to be statistically meaningful, paths between observations should preferentially traverse high-likelihood regions. Existing scalable pullback constructions typically use a unimodal Gaussian latent distribution, assuming that the data reside close to a single manifold. For multimodal data, mapping separated modes or local structures into one Gaussian region can require substantial transport deformation and compromise the resulting geometry. We introduce a pullback geometry for data supported on mixtures of manifolds. Using a latent Gaussian mixture, we define its Riemannian metric as the matrix square of the responsibility-weighted expected component precision. The metric is smooth and positive definite and recovers the existing Gaussian construction in the single-component limit. For structured overlapping mixtures, we establish conditions under which the log-density is concave along geodesics, providing a formal connection between the proposed geometry and paths through high-likelihood regions, and derive the corresponding local curvature relations. We instantiate this geometry in a normalizing flow with adaptive mixture learning, allowing the number of active components to emerge from the data and supporting component-wise reconstruction and local effective-dimension estimation. Experiments on synthetic geometric data, a controlled multi-view image setting with a known reference trajectory, and MNIST show reduced transport distortion, competitive path support, close reference-trajectory recovery, and improved interpolation realism. These results extend scalable pullback geometry beyond datasets that reside close to a single manifold while retaining tractable and interpretable local structure.

0.28

Minimax Optimality of the NPMLE for Gaussian Mixtures in Hellinger Distance

cs.IT math.STcs.LG ·2026-10-21 ·Shayan Hundrieser, Tudor Manole
similarity 0.56
Abstract

We analyze the problem of estimating univariate Gaussian mixture densities using the nonparametric maximum likelihood estimator (NPMLE) based on $n$ independent and identically distributed samples. We show that the unconstrained NPMLE achieves the minimax optimal rate $\log(n)/n$ in squared Hellinger distance, when the true mixing measure is sub-Gaussian. In addition, we show that when the true mixing measure has compact support, the NMPLE constrained to a compact interval also achieves the minimax optimal rate, which is slightly faster and of order $\log(n)/(n \log \log n)$. Our proof consists of sharply characterizing the local Hellinger bracketing entropy of the class of finite Gaussian mixtures. One of our key technical innovations is to sharply bound the $L^2$ bracketing entropy of balls of bounded-degree polynomials, which may be of independent interest. As an application, we establish the regret-optimality of $g$-modeling procedures based on the NPMLE in the Gaussian Empirical Bayes problem with bounded priors.

0.30

A Gamma-Mixture Solution for Definite Noncentral Complex Gaussian Quadratic Forms

eess.SP cs.IT ·2026-11-05 ·Fernando D. Almeida García, Carlos R. Nogueira da Silva, Felipe A. P. de Figueiredo, Rausley A. A. de Souza, Michel Daoud Yacoub +1 more
unusual similarity 0.56
Abstract

Finding exact solutions for the statistics of definite noncentral complex Gaussian quadratic forms (CGQFs) has long been a challenging problem. Among existing exact approaches, Raphaeli's formulation [1] remains the most computationally efficient. However, its analytical tractability is hindered by the dependence of its series coefficients on the evaluation variable, which confines their statistics to the probability density function (PDF) and cumulative distribution function (CDF). This paper overcomes this limitation by deriving novel Gamma-mixture representations for the PDF and CDF of positive- and negative-definite noncentral CGQFs. The resulting structure broadens Raphaeli's statistical framework beyond the PDF and CDF to a wider class of CGQF functionals and a substantial range of practical applications. In particular, expectations of integrable functions of definite noncentral CGQFs are expressed as weighted sums of the corresponding expectations over Gamma random variables, enabling the direct derivation of moments, transforms, and performance measures. Moreover, the proposed formulation reduces the single-point complexity from \(O\!\left(n^2K+nK^2\right)\) for Raphaeli's method to \(O\!\left(nK+K^2\right)\), where \(n\) denotes the quadratic-form dimension and \(K\) the truncation order. The proposed Gamma-mixture structure also enables closed-form approximations that, by construction, remain valid probability distributions for any truncation order. In addition, uniform PDF and CDF truncation-error bounds are derived, allowing prescribed-accuracy truncation. As a sample application, we derive the first reported exact average bit error rate expression for maximal-ratio combining receivers with an arbitrary number of branches under an arbitrary positive-definite Rician correlation structure.

0.28

Bregman geometry in regular exponential families: dual representations and relative smoothness

cs.LG stat.MLmath.OC ·2026-11-13 ·Mame Diarra Fall, Éric Barat, Nicolas Dobigeon
similarity 0.55
Abstract

Regular exponential-family distributions are ubiquitous in statistics due to their appealing and convenient properties. Notably, they can be expressed in terms of Bregman divergences, revealing fundamental connections between statistics, convex analysis, and information geometry. However, this geometric structure has not been systematically exploited to characterize these distributions with respect to their expectation parametrization. Building on a projective characterization, we establish a general construction of equivalent dual Bregman representations, together with normalization and uniqueness results. These representations express log-densities as ordinary Bregman divergences in dual coordinate systems, providing geometric characterizations of the parameter space. So-called $β$- and $α$-scaled dual representations emerge as particular instances of this framework. We further derive dual Bregman representations of conjugate priors and establish that they preserve conjugacy. Applications to classical distributions illustrate the geometric structures underlying seemingly different statistical models. Moreover, we show that these representations provide natural reference geometries for establishing relative smoothness of the primal Bregman representation of the distribution, with explicit bounds that are sharp in the scalar and separable settings. An illustrative application to the Poisson distribution sheds new light on several works in the literature, providing complementary geometric justifications for reference geometries while deriving existing algorithms.

0.27

Dimension-free comparison inequalities between $f$-divergences for Gaussian location mixtures

cs.IT math.STmath.PRstat.ML ·2026-10-18 ·Jeonghwan Lee, Cong Ma
similarity 0.55
Abstract · 21 pages; Comments would be appreciated!

We establish dimension-free comparison inequalities between the Kullback-Leibler (KL) and $χ^2$-divergences and the squared Hellinger distance for Gaussian location mixtures with compactly supported mixing distributions. We show that both divergences are bounded by the squared Hellinger distance up to constants depending only on the radius of the support, independently of the ambient dimension and the number of mixture components. These results resolve an open question raised by \cite{jia2023entropic} and strengthen existing comparison inequalities. Our proof relies on a Gaussian interpolation argument that relates the derivatives of these divergences to differences between posterior means.

0.29