Please click on the right menu to get the abstracts.
April 9, (Fri) 2021 - Chair Rosanna Campagna
17:20--17:30 Welcome and virtual coffee
17:30--18:00 Brigitte Forster, Frame Recycling
18:00--18:30 Filomena Di Tommaso, Multinode Shepard methods and Applications
Brigitte Forster, Frame Recycling
Joint work with: Peter Massopust (TU Munchen), Ole Christensen (DTU Lyngby) and Florian Heinrich (University of Passau).
Abstract: Grafakos and Sansing have shown how to obtain directionally sensitive time-frequency decompositions in L^2(R^n) based on Gabor systems in L^2(R); the key tool is the "ridge idea", which lifts a function of one variable to a function of several variables. We extend their idea by showing that similar results hold starting with general frames for L^2(R), both in the setting of discrete frames and continuous frames. This allows to apply the theory for several other classes of frames, e.g., wavelet frames and shift-invariant systems. We will consider applications to the Meyer wavelet and the B-spline family. We will close with a short discussion of partial ridges.
Filomena Di Tommaso, Multinode Shepard methods and Applications
Abstract: In this talk we give an overview on the Multinode Shepard methods for scattered data interpolation and apply them to the numerical solution of PDEs by collocation. These methods derive from the Little’s observation about the possibility to improve the precision and the behaviour of the classical Shepard interpolants only by fixing triangulations of the scattered nodes and by blending local linear interpolants on the vertices of triangles with Shepard like basis functions based on those triangles. The main advantage of the Little interpolants is their explicit expressions that rely only on function values without using any derivative data (exact or approximated). Moreover, it is possible to provide algorithms for their fast computation based on a criterion of choice of triangles (which may overlap or being disjoint) and on a searching technique to detect and select the nearest neighbour points. Further improvements of such interpolants require the solution of two main problems: the partitioning of the node set in ordered subsets that guarantees the existence and accuracy of approximation of local interpolation polynomials of fixed total degree and the possibility to compute them in a stable way. In line with the Kansa idea, the multinode Shepard operators can be applied in the numerical solution of elliptic PDEs with Dirichlet conditions via collocation. Numerical experiments show accuracy of approximation comparable with that one of the Kansa method but a lower condition number of the collocation matrix.
April 12, (Mon) 2021 - Chair Costanza Conti
17:30--18:00 Tom Lyche, B-spline like bases for the Clough-Tocher split based on simplex splines
18:00--18:30 Nir Sharon, Pyramid transform for manifold-valued data
Tom Lyche, B-spline like bases for the Clough-Tocher split based on simplex splines
Abstract: Piecewise polynomials, orsplinesdefined over triangulations, have applications inmany branches of science. In applications like geometric modelling and solving PDEs by isogeometric methods one often desires a low degree spline with higher smoothness.To compute with splines we need a basis for the spline space. In the univariate case B-splines have many advantages. They lead to banded matrices with good stability properties for low degrees and can be computed efficiently using stable recurrence relations. It would be nice to have bases with similar properties fors plines on triangulations. In [1] a basis of simplex splines was introduced for C^1quadratics on a split where each triangle is divided into 12 sub-triangles by connecting edge midpoints and vertices (the PS-12 split). This basis has all the usual properties of univariate B-splines, including a recurrence relation down to piecewise linear polynomials and a Marsden identity. Global C1-smoothness is achieved by connecting neighbooring triangles using classical Bézier techniques. This construction was extended to C^2-quintics in [2, 3]. In this talk we give an introduction to simplex splines and consider in detail applications to smooth splines on the Clough-Tocher split, where each triangle is divided into three subtriangles by connecting the vertices to the barycenter. This is joint work with Jean-Louis Merrien and Tomas Sauer.
[1] E. Cohen, T. Lyche, and R.F. Riesenfeld, A B-spline-like basis for the Powell-Sabin 12-split based on simplex splines, Matematics of Computation 82, 1667–1707 (2013).
[2] T. Lyche and G. Muntingh, A Hermite interpolatory subdivision scheme for C^2-quintics onthe Powell–Sabin 12-split, Comput. Aided Geom. Design, 31(2014), 464–474.
[3] T. Lyche and G. Muntingh,Stable Simplex Spline Bases for C^3 Quintics on the Powell–Sabin 12-Split, Constr. Approx.,45(2017), 1–32.
Nir Sharon, Pyramid transform for manifold-valued data
Abstract: A multiscale transform is a standard tool in signal and image processing that enables a hierarchical analysis. In the talk, we introduce a multiscale pyramid transform for analyzing manifold-valued functions. The transform is based upon a unique class of downsampling operators that enables non-interpolatory subdivision schemes as upsampling operators. We describe this construction in detail and present its analytical properties, including stability and coefficient decay. We numerically demonstrate the results and show the application of our method for denoising and abnormalities detection.
April 20, (Tue) 2021 - Chair Emma Perracchione
17:30--18:00 Elisabeth Larsson, The least-squares RBF-FD method
18:00--18:30 Sourav Dutta, Kernel-based approximation methods for reduced order modeling applications in hydrology
Elisabeth Larsson, The least-squares RBF-FD method
Joint work with: Igor Tominec (Divison of Scientific Computing, Department of Information Technology, Uppsala University, Sweden) and Alfa Heryudono (Department of Mathematics, University of Massachusetts Dartmouth, MA, USA.)
Abstract: The radial basis function-generated finite difference method (RBF-FD) aims to combine the simplicity of finite difference methods and the flexibility of finite element methods. The approximations are based on scattered data, and do not require a grid or a mesh. Stencil weights for approximation of function values and derivatives are computed based on interpolation of local data. The RBF-FD method was first introduced as a collocation method, where partial differential equations (PDEs) and boundary conditions are enforced at the stencil points. Here, we introduce the least-squares RBF-FD method, where the PDE and boundary conditions are enforced at M>N points, where N is the number of stencil points. The introduction of oversampling improves the stability of the approximation as well as its behaviour close to boundaries with Neumann conditions. This new formulation of the method also allowed us to prove theoretical convergence results for a Poisson problem with mixed boundary conditions. In the talk, we discuss the principles behind the convergence results, we show numerical experiments that support the theory, as well as results from some non-trivial application problems.
Sourav Dutta, Kernel-based approximation methods for reduced order modeling applications in hydrology
Abstract: The depth-averaged two-dimensional Shallow Water Equations (SWE) are a well-studied system of hyperbolic or near-hyperbolic partial differential equations (PDEs) that are widely adopted to study various flow scenarios which arise in engineering applications that are critical for achieving the U.S. Army Corps of Engineer’s mission of delivering vital public and military engineering services.
However, for multi-query, real-time and fast-replay applications arising in optimal design, risk assessment, or ensemble forecasting problems, fully resolved numerical simulations of a parametrized SWE model pose a significant computational challenge. Reduced order modeling (ROM) encompasses a wide array of approaches that aim to drastically reduce the computational burden versus reference, high-fidelity simulations. These techniques introduce a linear or non-linear mapping (or projection) from the space of high-fidelity (full order) solutions to a lower-dimensional latent space in which the model dynamics can be efficiently approximated using various data-driven methods.
In this talk, radial basis functions (RBF) will be adopted as a classical kernel-based regression method to devise non-intrusive latent space representations of the high-fidelity model dynamics. A novel greedy strategy will be presented to generate an optimal selection of RBF centers. Moreover, higher-order discretization techniques will be employed to achieve improved stability in the long-term forecasting of the system coefficients in the latent space. The performance of this framework will be demonstrated using real-world, large-scale riverine and coastal flow problems. Quantitative comparison results with existing data-driven strategies like Gaussian process regression (GPR), dynamic mode decomposition (DMD), and machine learning (ML)-based time series learning techniques will be presented to assess predictive accuracy, efficiency, and computational stability of the different reduced order models.
April 28, (Wed) 2021 - Chair Virginie Uhlmann
17:30--18:00 Demetrio Labate, Analysis of the image inpainting problem using sparse multiscale representation
18:00--18:30 Anaïs Badoual, Active Subdivision Surfaces for the Semiautomatic Segmentation of Biomedical Volumes
Demetrio Labate, Analysis of the image inpainting problem using sparse multiscale representation
Abstract: Image inpainting is an imaging processing task aiming at recovering missing blocks of data in an image or a video. Sparse multiscale representations offer both an efficient algorithmic framework and well justified theoretical setting to address the image inpainting problem. In this talk, I will formulate this task in the continuous domain as a function interpolation problem in a Hilbert space. Under the assumption that the unknown image is sparse with respect to an appropriate representation, the solution of the inpainting problem is also the sparsest admissible solution. As images found in many applications are dominated by edges, I will assume a simplified image model consisting of distributions supported on curvilinear singularities. I will prove that the theoretical performance of image inpainting depends on the microlocal properties of the representation system, namely, exact image recovery is achieved if the size of the missing singularity is smaller than the size of the structure elements of the representation system. A consequence of this observation is that a shearlet-based image inpainting algorithm - exploiting their sparser approximation properties - significantly outperforms a similar approach based on traditional multiscale methods.
Anaïs Badoual, Active Subdivision Surfaces for the Semiautomatic Segmentation of Biomedical Volumes
Abstract: We present a new family of active surfaces for the semiautomatic segmentation of volumetric objects in 3D biomedical images. We represent our deformable model by a subdivision surface encoded by a small set of control points and generated through a geometric refinement process. The subdivision operator confers important properties to the surface such as smoothness, reproduction of desirable shapes and interpolation of the control points. We deform the subdivision surface through the minimization of suitable gradient-based and region-based energy terms that we have designed for that purpose. In addition, we provide an easy way to combine these energies with convolutional neural networks. Our active subdivision surface satisfies the property of multiresolution, which allows us to adopt a coarse-to fine optimization strategy. This speeds up the computations and decreases its dependence on initialization compared to singleresolution active surfaces. Performance evaluations on both synthetic and real biomedical data show that our active subdivision surface is robust in the presence of noise and outperforms current state-of-the-art methods. In addition, we provide a software that gives full control over the active subdivision.