Gaussian mixture model decomposition of multivariate signals

Abstract

We propose a greedy variational method for decomposing a non-negative multivariate signal as a weighted sum of Gaussians, which, borrowing the terminology from statistics, we refer to as a Gaussian mixture model. Notably, our method has the following features: (1) It accepts multivariate signals, i.e., sampled multivariate functions, histograms, time series, images, etc., as input. (2) The method can handle general (i.e., ellipsoidal) Gaussians. (3) No prior assumption on the number of mixture components is needed. To the best of our knowledge, no previous method for Gaussian mixture model decomposition simultaneously enjoys all these features. We also prove an upper bound, which cannot be improved by a global constant, for the distance from any mode of a Gaussian mixture model to the set of corresponding means. For mixtures of spherical Gaussians with common variance $$\sigma ^2$$ σ 2 , the bound takes the simple form $$\sqrt{n}\sigma $$ n σ . We evaluate our method on one- and two-dimensional signals. Finally, we discuss the relation between clustering and signal decomposition, and compare our method to the baseline expectation maximization algorithm.

Metal artefact reduction in digital tomosynthesis using component decomposition

“…Usually, decomposition of a multimodal distribution is performed by fitting components, and representing each one by its mean and the standard deviation, i.e. components are assumed to be Gaussians [18][19][20]. Although this works in many cases, here, however, we are looking for a more generic approach.…”

Localización simultánea y mapeo para control de un robot móvil autónomo usando escaneo de nube de puntos LiDAR y métodos de aprendizaje de máquina

“…La novedad de este trabajo radica en implementar métodos de aprendizaje no supervisado como: K-means [20], GMM [21], y DBSCAN [22], para obtener de forma simple un mapa probabilístico del entorno navegable del robot en base a lecturas de rango de un sensor LiDAR. Este trabajo contribuye con una estrategia integral de SLAM que fusiona los métodos de aprendizaje no supervisado propuestos con EKF para la localización simultánea de un robot autónomo que navega en tiempo real [23].…”

Joint Gaussian dictionary learning and tomographic reconstruction

“…First, each projection data g j is decomposed into an n − 1 dimensional GMM g G := D ( j) , α ( j) m j=1 . More precisely, we use a method from [ZY21] for evaluating C S ϕ ⊥ to compute…”

“…For g G no pre-processing was necessary, since it was already in GMM form. For g and g noisy we used the method of [ZY21] for the pre-processing step (line 2 of algorithm 1). We used the L-BFGS-B method [BLNZ95] for all minimization problems in algorithms 1 and 2.…”