arXiv subjects

Felix Breuer

Publications and source records attributed to Felix Breuer.

At least 19 recordsLinked to original sources

arXiv2025

Controlling spatial correlation in k-space interpolation networks for MRI reconstruction: denoising versus apparent blurring

Purpose: Interpretability is essential for the clinical adoption of state-of-the-art machine learning (ML) methods in magnetic resonance imaging (MRI). Conventional evaluation of ML reconstructions relies heavily on aggregate image metrics that require fully sampled references. These metrics, inherited from classical image processing and natural image ML, often overlook the critical challenge of noise amplification specific to medical image reconstruction. This study aims to analyze the influence of nonlinear activations on spatial noise variance distribution of k-space interpolation networks (RAKI) and to provide a framework for incorporating variance maps during network training. Methods: We present an analytical framework that decomposes pixel-level noise variance into components reflecting linear and nonlinear characteristics of RAKI. By applying automatic differentiation on the image-space equivalent of the network, variance maps are computed during each training iteration, enabling runtime quality assessment beyond data consistency. We introduce apparent blurring, quantifying nonlinear signal mixing without dependence on reference images. By incorporating variance maps into the traning loss as regularizers, our self-informed RAKI architecture (G-factor-informed RAKI, GIF-RAKI) can directly integrate updated noise characteristics during runtime. Results: Experimental results demonstrate that variance components quantitatively explain network behavior. GIF-RAKI outperforms conventional RAKI variants in image fidelity and noise suppression. Conclusion: Our methodology advances practical and theoretical aspects of ML-based MRI reconstruction by reinstating reconstruction noise characterization as a cornerstone for performance evaluation, eliminating the need for fully sampled references. GIF-RAKI also enables optimization of the trade-off between denoising and apparent blurring.

physics.med-ph↗

arXiv2025

The Arboricity Polynomial

We introduce a new matroid (graph) invariant, the arboricity polynomial. Given a matroid, the arboricity polynomial enumerates the number of covers of the ground set by disjoint independent sets. We establish the polynomiality of the counting function as a special case of a scheduling polynomial, i.e. both in terms of quasisymmetric functions and via Ehrhart theory of the normal fan of the matroid base polytope. We show basic properties of the polynomial and demonstrate that it is not a Tutte invariant. Namely, the arboricity polynomial does not satisfy a contraction / deletion recursion.

math.CO↗

arXiv2024

Image space formalism of convolutional neural networks for k-space interpolation

Purpose: Noise resilience in image reconstructions by scan-specific robust artificial neural networks for k-space interpolation (RAKI) is linked to nonlinear activations in k-space. To gain a deeper understanding of this relationship, an image space formalism of RAKI is introduced for analyzing noise propagation analytically, identifying and characterizing image reconstruction features and to describe the role of nonlinear activations in a human readable manner. Methods: The image space formalism for RAKI inference is employed by expressing nonlinear activations in k-space as element-wise multiplications with activation masks, which transform into convolutions in image space. Jacobians of the de-aliased, coil-combined image relative to the aliased coil images can be expressed algebraically, and thus, the noise amplification is quantified analytically (g-factor maps). We analyze the role of nonlinearity for noise resilience by controlling the degree of nonlinearity in the reconstruction model via the negative slope parameter in leaky ReLU. Results: The analytical g-factor maps correspond with those obtained from Monte Carlo simulations and from an auto differentiation approach for in vivo brain images. Apparent blurring and contrast loss artifacts are identified as implications of enhanced noise resilience. These residual artifacts can be traded against noise resilience by adjusting the degree of nonlinearity in the model (Tikhonov-like regularization) in case of limited training data. The inspection of image space activations reveals an autocorrelation pattern leading to a potential center artifact. Conclusion: The image space formalism of RAKI provides the means for analytical quantitative noisepropagation analysis and human-readable visualization of the effects of the nonlinear activation functions in k-space.

cs.CV↗

arXiv2022

Iterative training of robust k-space interpolation networks for improved image reconstruction with limited scan specific training samples

Purpose: To evaluate an iterative learning approach for enhanced performance of Robust Artificial-neural-networks for K-space Interpolation (RAKI), when only a limited amount of training data (auto-calibration signals, ACS) are available for accelerated standard 2D imaging. Methods: In a first step, the RAKI model was optimized for the case of strongly limited training data amount. In the iterative learning approach (termed iterative RAKI), the optimized RAKI model is initially trained using original and augmented ACS obtained from a linear parallel imaging reconstruction. Subsequently, the RAKI convolution filters are refined iteratively using original and augmented ACS extracted from the previous RAKI reconstruction. Evaluation was carried out on 200 retrospectively undersampled in-vivo datasets from the fastMRI neuro database with different contrast settings. Results: For limited training data (18 and 22 ACS lines for R=4 and R=5, respectively), iterative RAKI outperforms standard RAKI by reducing residual artefacts and yields strong noise suppression when compared to standard parallel imaging, underlined by quantitative reconstruction quality metrics. In combination with a phase constraint, further reconstruction improvements can be achieved. Additionally, iterative RAKI shows better performance than both GRAPPA and RAKI in case of pre-scan calibration with varying contrast between training- and undersampled data. Conclusion: The iterative learning approach with RAKI benefits from standard RAKIs well known noise suppression feature but requires less original training data for the accurate reconstruction of standard 2D images thereby improving net acceleration.

eess.IV↗

arXiv2015

Scheduling Problems

We introduce the notion of a scheduling problem which is a boolean function $S$ over atomic formulas of the form $x_i \leq x_j$. Considering the $x_i$ as jobs to be performed, an integer assignment satisfying $S$ schedules the jobs subject to the constraints of the atomic formulas. The scheduling counting function counts the number of solutions to $S$. We prove that this counting function is a polynomial in the number of time slots allowed. Scheduling polynomials include the chromatic polynomial of a graph, the zeta polynomial of a lattice, the Billera-Jia-Reiner polynomial of a matroid. To any scheduling problem, we associate not only a counting function for solutions, but also a quasisymmetric function and a quasisymmetric function in non-commuting variables. These scheduling functions include the chromatic symmetric functions of Sagan, Gebhard, and Stanley, and a close variant of Ehrenborg's quasisymmetric function for posets. Geometrically, we consider the space of all solutions to a given scheduling problem. We extend a result of Steingrímmson by proving that the $h$-vector of the space of solutions is given by a shift of the scheduling polynomial. Furthermore, under certain niceness conditions on the defining boolean function, we prove partitionability of the space of solutions and positivity of fundamental expansions of the scheduling quasisymmetric functions and of the $h$-vector of the scheduling polynomial.

math.CO↗

arXiv2015

Polyhedral geometry, supercranks, and combinatorial witnesses of congruences for partitions into three parts

In this paper, we use a branch of polyhedral geometry, Ehrhart theory, to expand our combinatorial understanding of congruences for partition functions. Ehrhart theory allows us to give a new decomposition of partitions, which in turn allows us to define statistics called {\it supercranks} that combinatorially witness every instance of divisibility of $p(n,3)$ by any prime $m \equiv -1 \pmod 6$, where $p(n,3)$ is the number of partitions of $n$ into three parts. A rearrangement of lattice points allows us to demonstrate with explicit bijections how to divide these sets of partitions into $m$ equinumerous classes. The behavior for primes $m' \equiv 1 \pmod 6$ is also discussed.

math.CO↗

arXiv2015

A polyhedral model of partitions with bounded differences and a bijective proof of a theorem of Andrews, Beck, and Robbins

The main result of this paper is a bijective proof showing that the generating function for partitions with bounded differences between largest and smallest part is a rational function. This result is similar to the closely related case of partitions with fixed differences between largest and smallest parts which has recently been studied through analytic methods by Andrews, Beck, and Robbins. Our approach is geometric: We model partitions with bounded differences as lattice points in an infinite union of polyhedral cones. Surprisingly, this infinite union tiles a single simplicial cone. This construction then leads to a bijection that can be interpreted on a purely combinatorial level.

math.CO↗

arXiv2015

Polyhedral Omega: A New Algorithm for Solving Linear Diophantine Systems

Polyhedral Omega is a new algorithm for solving linear Diophantine systems (LDS), i.e., for computing a multivariate rational function representation of the set of all non-negative integer solutions to a system of linear equations and inequalities. Polyhedral Omega combines methods from partition analysis with methods from polyhedral geometry. In particular, we combine MacMahon's iterative approach based on the Omega operator and explicit formulas for its evaluation with geometric tools such as Brion decompositions and Barvinok's short rational function representations. In this way, we connect two recent branches of research that have so far remained separate, unified by the concept of symbolic cones which we introduce. The resulting LDS solver Polyhedral Omega is significantly faster than previous solvers based on partition analysis and it is competitive with state-of-the-art LDS solvers based on geometric methods. Most importantly, this synthesis of ideas makes Polyhedral Omega the simplest algorithm for solving linear Diophantine systems available to date. Moreover, we provide an illustrated geometric interpretation of partition analysis, with the aim of making ideas from both areas accessible to readers from a wide range of backgrounds.

math.CO↗

arXiv2014

An Invitation to Ehrhart Theory: Polyhedral Geometry and its Applications in Enumerative Combinatorics

In this expository article we give an introduction to Ehrhart theory, i.e., the theory of integer points in polyhedra, and take a tour through its applications in enumerative combinatorics. Topics include geometric modeling in combinatorics, Ehrhart's method for proving that a couting function is a polynomial, the connection between polyhedral cones, rational functions and quasisymmetric functions, methods for bounding coefficients, combinatorial reciprocity theorems, algorithms for counting integer points in polyhedra and computing rational function representations, as well as visualizations of the greatest common divisor and the Euclidean algorithm.

math.CO↗

arXiv2013

Enumerating Colorings, Tensions and Flows in Cell Complexes

We study quasipolynomials enumerating proper colorings, nowhere-zero tensions, and nowhere-zero flows in an arbitrary CW-complex $X$, generalizing the chromatic, tension and flow polynomials of a graph. Our colorings, tensions and flows may be either modular (with values in $\mathbb{Z}/k\mathbb{Z}$ for some $k$) or integral (with values in $\{-k+1,\dots,k-1\}$). We obtain deletion-contraction recurrences and closed formulas for the chromatic, tension and flow quasipolynomials, assuming certain unimodularity conditions. We use geometric methods, specifically Ehrhart theory and inside-out polytopes, to obtain reciprocity theorems for all of the aforementioned quasipolynomials, giving combinatorial interpretations of their values at negative integers as well as formulas for the numbers of acyclic and totally cyclic orientations of $X$.

math.CO↗

arXiv2013

Analytic Feature Selection for Support Vector Machines

Support vector machines (SVMs) rely on the inherent geometry of a data set to classify training data. Because of this, we believe SVMs are an excellent candidate to guide the development of an analytic feature selection algorithm, as opposed to the more commonly used heuristic methods. We propose a filter-based feature selection algorithm based on the inherent geometry of a feature set. Through observation, we identified six geometric properties that differ between optimal and suboptimal feature sets, and have statistically significant correlations to classifier performance. Our algorithm is based on logistic and linear regression models using these six geometric properties as predictor variables. The proposed algorithm achieves excellent results on high dimensional text data sets, with features that can be organized into a handful of feature types; for example, unigrams, bigrams or semantic structural features. We believe this algorithm is a novel and effective approach to solving the feature selection problem for linear SVMs.

cs.LG↗

arXiv2012

Hypergraph Coloring Complexes

The aim of this paper is to generalize the notion of the coloring complex of a graph to hypergraphs. We present three different interpretations of those complexes -- a purely combinatorial one and two geometric ones. It is shown, that most of the properties, which are known to be true for coloring complexes of graphs, break down in this more general setting, e.g., Cohen-Macaulayness and partitionabilty. Nevertheless, we are able to provide bounds for the $f$- and $h$-vectors of those complexes which yield new bounds on chromatic polynomials of hypergraphs. Moreover, it is shown that the coloring complex of a hypergraph has a wedge decomposition, though we conjecture that in general this decomposition is not homotopy equivalent to a wedge of spheres. In addition, we can completely characterize those hypergraphs whose coloring complex is connected.

math.CO↗

arXiv2012

Ehrhart f*-coefficients of polytopal complexes are non-negative integers

The Ehrhart polynomial $L_P$ of an integral polytope $P$ counts the number of integer points in integral dilates of $P$. Ehrhart polynomials of polytopes are often described in terms of their Ehrhart $h^*$-vector (aka Ehrhart $δ$-vector), which is the vector of coefficients of $L_P$ with respect to a certain binomial basis and which coincides with the $h$-vector of a regular unimodular triangulation of $P$ (if one exists). One important result by Stanley about $h^*$-vectors of polytopes is that their entries are always non-negative. However, recent combinatorial applications of Ehrhart theory give rise to polytopal complexes with $h^*$-vectors that have negative entries. In this article we introduce the Ehrhart $f^*$-vector of polytopes or, more generally, of polytopal complexes $K$. These are again coefficient vectors of $L_K$ with respect to a certain binomial basis of the space of polynomials and they have the property that the $f^*$-vector of a unimodular simplicial complex coincides with its $f$-vector. The main result of this article is a counting interpretation for the $f^*$-coefficients which implies that $f^*$-coefficients of integral polytopal complexes are always non-negative integers. This holds even if the polytopal complex does not have a unimodular triangulation and if its $h^*$-vector does have negative entries. Our main technical tool is a new partition of the set of lattice points in a simplicial cone into discrete cones. Further results include a complete characterization of Ehrhart polynomials of integral partial polytopal complexes and a non-negativity theorem for the $f^*$-vectors of rational polytopal complexes.

math.CO↗

arXiv2011

Image acceleration in parallel magnetic resonance imaging by means of metamaterial magnetoinductive lenses

Parallel magnetic resonance imaging (MRI) is a technique of image acceleration which takes advantage of the localization of the field of view (FOV) of coils in an array. In this letter we show that metamaterial lenses based on capacitively-loaded rings can provide higher localization of the FOV. Several lens designs are systematically analyzed in order to find the structure providing higher signal-to-noise-ratio. The magnetoinductive (MI) lens is find to be the optimum structure and an experiment is developed to show it. The ability of the fabricated MI lenses to accelerate the image is quantified by means of the parameter known in the MRI community as g-factor.

physics.med-ph↗

arXiv2010

Bounds on the Coefficients of Tension and Flow Polynomials

The goal of this article is to obtain bounds on the coefficients of modular and integral flow and tension polynomials of graphs. To this end we make use of the fact that these polynomials can be realized as Ehrhart polynomials of inside-out polytopes. Inside-out polytopes come with an associated relative polytopal complex and, for a wide class of inside-out polytopes, we show that this complex has a convex ear decomposition. This leads to the desired bounds on the coefficients of these polynomials.

math.CO↗

arXiv2009

Viewing counting polynomials as Hilbert functions via Ehrhart theory

Steingrimsson (2001) showed that the chromatic polynomial of a graph is the Hilbert function of a relative Stanley-Reisner ideal. We approach this result from the point of view of Ehrhart theory and give a sufficient criterion for when the Ehrhart polynomial of a given relative polytopal complex is a Hilbert function in Steingrimsson's sense. We use this result to establish that the modular and integral flow and tension polynomials of a graph are Hilbert functions.

math.CO↗

arXiv2009

Counting Hexagonal Patches and Independent Sets in Circle Graphs

A hexagonal patch is a plane graph in which inner faces have length 6, inner vertices have degree 3, and boundary vertices have degree 2 or 3. We consider the following counting problem: given a sequence of twos and threes, how many hexagonal patches exist with this degree sequence along the outer face? This problem is motivated by the study of benzenoid hydrocarbons and fullerenes in computational chemistry. We give the first polynomial time algorithm for this problem. We show that it can be reduced to counting maximum independent sets in circle graphs, and give a simple and fast algorithm for this problem.

cs.DM↗

arXiv2009

Finding Fullerene Patches in Polynomial Time

We consider the following question, motivated by the enumeration of fullerenes. A fullerene patch is a 2-connected plane graph G in which inner faces have length 5 or 6, non-boundary vertices have degree 3, and boundary vertices have degree 2 or 3. The degree sequence along the boundary is called the boundary code of G. We show that the question whether a given sequence S is a boundary code of some fullerene patch can be answered in polynomial time when such patches have at most five 5-faces. We conjecture that our algorithm gives the correct answer for any number of 5-faces, and sketch how to extend the algorithm to the problem of counting the number of different patches with a given boundary code.

cs.DM↗