SearcharxivSearch

arXiv subjects

Christof Puhle

Publications and source records attributed to Christof Puhle.

7 recordsLinked to original sources

Deep learning-based filtering of cross-spectral matrices using generative adversarial networks

In this paper, we present a deep-learning method to filter out effects such as ambient noise, reflections, or source directivity from microphone array data represented as cross-spectral matrices. Specifically, we focus on a generative adversarial network (GAN) architecture designed to transform fixed-size cross-spectral matrices. Theses models were trained using sound pressure simulations of varying complexity developed for this purpose. Based on the results from applying these methods in a hyperparameter optimization of an auto-encoding task, we trained the optimized model to perform five distinct transformation tasks derived from different complexities inherent in our sound pressure simulations.

cs.SD

On generalized quasi-Sasaki manifolds

We study 5-dimensional Riemannian manifolds that admit an almost contact metric structure. In particular, we generalize the class of quasi-Sasaki manifolds and characterize these structures by their intrinsic torsion. Among other things, we see that these manifolds admit a unique metric connection that is compatible with the underlying almost contact metric structure. Finally, we construct a family of examples that are not quasi-Sasaki.

math.DG

Almost contact metric 5-manifolds and connections with torsion

We study 5-dimensional Riemannian manifolds that admit an almost contact metric structure. We classify these structures by their intrinsic torsion and review the literature in terms of this scheme. Moreover, we determine necessary and sufficient conditions for the existence of metric connections with vectorial, totally skew-symmetric or traceless cyclic torsion that are compatible with the almost contact metric structure. Finally, we examine explicit examples of almost contact metric 5-manifolds from this perspective.

math.DG

Riemannian manifolds with structure group PSU(3)

We study 8-dimensional Riemannian manifolds that admit a PSU(3)-structure. We classify these structures by their intrinsic torsion and characterize the corresponding classes via differential equations. Moreover, we consider a connection defined by a 3- and a 4-form that preserves the underlying structure. Finally, we discuss the geometry of these manifolds relatively to the holonomy algebra of this connection.

math.DG

Spin(7)-manifolds with parallel torsion form

Any Spin(7)-manifold admits a metric connection \nabla^c with totally skew-symmetric torsion T^c preserving the underlying structure. We classify those with \nabla^c-parallel T^c\neq0 and non-Abelian isotropy algebra iso(T^c)<spin(7). These are isometric to either Riemannian products or homogeneous naturally reductive spaces, each admitting two \nabla^c-parallel spinor fields.

math.DG

The Killing spinor equation with higher order potentials

Let (M^n,g) be a Riemannian spin manifold. The basic equations in supergravity models of type IIa string theory with 4-form flux involve a 3-form T, a 4-form F, a spinorial covariant derivative \nabla depending on \nabla^g, T, F, and a \nabla-parallel spinor field Ψ. We classify and construct many explicit families of solutions to this system of spinorial field equations by means of non-integrable special geometries. The latter include α-Sasakian structures in dimensions 5 and 7, almost Hermitian structures in dimension 6 and cocalibrated G_2-structures in dimension 7. We show that there are several examples also satisfying an additional constraint for the energy-momentum tensor.

math.DG

Remarks on "Resolving isospectral `drums' by counting nodal domains"

In [3] the authors studied the 4-parameter family of isospectral flat 4-tori T^\pm(a,b,c,d) discovered by Conway and Sloane. With a particular method of counting nodal domains they were able to distinguish these tori (numerically) by computing the corresponding nodal sequences relative to a few explicit tuples (a,b,c,d). In this note we confirm the expectation expressed in [3] by proving analytically that their nodal count distinguishes any 4-tuple of distinct positive real numbers.

math.SP