Searcharxiv⌕ Search

arXiv subjects

Anton Savostianov

Publications and source records attributed to Anton Savostianov.

16 recordsLinked to original sources
arXiv2025

Grassmanian Interpolation of Low-Pass Graph Filters: Theory and Applications

Low-pass graph filters are fundamental for signal processing on graphs and other non-Euclidean domains. However, the computation of such filters for parametric graph families can be prohibitively expensive as computation of the corresponding low-frequency subspaces, requires the repeated solution of an eigenvalue problem. We suggest a novel algorithm of low-pass graph filter interpolation based on Riemannian interpolation in normal coordinates on the Grassmann manifold. We derive an error bound estimate for the subspace interpolation and suggest two possible applications for induced parametric graph families. First, we argue that the temporal evolution of the node features may be translated to the evolving graph topology via a similarity correction to adjust the homophily degree of the network. Second, we suggest a dot product graph family induced by a given static graph which allows to infer improved message passing scheme for node classification facilitated by the filter interpolation.

cs.LG↗
arXiv2025

Efficient Sparsification of Simplicial Complexes via Local Densities of States

Simplicial complexes (SCs) have become a popular abstraction for analyzing complex data using tools from topological data analysis or topological signal processing. However, the analysis of many real-world datasets often leads to dense SCs, with many higher-order simplicies, which results in prohibitive computational requirements in terms of time and memory consumption. The sparsification of such complexes is thus of broad interest, i.e., the approximation of an original SC with a sparser surrogate SC (with typically only a log-linear number of simplices) that maintains the spectrum of the original SC as closely as possible. In this work, we develop a novel method for a probabilistic sparsification of SCs that uses so-called local densities of states. Using this local densities of states, we can efficiently approximate so-called generalized effective resistance of each simplex, which is proportional to the required sampling probability for the sparsification of the SC. To avoid degenerate structures in the spectrum of the corresponding Hodge Laplacian operators, we suggest a ``kernel-ignoring'' decomposition to approximate the sampling probability. Additionally, we utilize certain error estimates to characterize the asymptotic algorithmic complexity of the developed method. We demonstrate the performance of our framework on a family of Vietoris--Rips filtered simplicial complexes.

stat.ML↗
arXiv2025

Convergence of gradient based training for linear Graph Neural Networks

Graph Neural Networks (GNNs) are powerful tools for addressing learning problems on graph structures, with a wide range of applications in molecular biology and social networks. However, the theoretical foundations underlying their empirical performance are not well understood. In this article, we examine the convergence of gradient dynamics in the training of linear GNNs. Specifically, we prove that the gradient flow training of a linear GNN with mean squared loss converges to the global minimum at an exponential rate. The convergence rate depends explicitly on the initial weights and the graph shift operator, which we validate on synthetic datasets from well-known graph models and real-world datasets. Furthermore, we discuss the gradient flow that minimizes the total weights at the global minimum. In addition to the gradient flow, we study the convergence of linear GNNs under gradient descent training, an iterative scheme viewed as a discretization of gradient flow.

cs.LG↗
arXiv2024

Contractivity of neural ODEs: an eigenvalue optimization problem

We propose a novel methodology to solve a key eigenvalue optimization problem which arises in the contractivity analysis of neural ODEs. When looking at contractivity properties of a one layer weight-tied neural ODE $\dot{u}(t)=σ(Au(t)+b)$ (with $u,b \in {\mathbb R}^n$, $A$ is a given $n \times n$ matrix, $σ: {\mathbb R} \to {\mathbb R}$ denotes an activation function and for a vector $z \in {\mathbb R}^n$, $σ(z) \in {\mathbb R}^n$ has to be interpreted entry-wise), we are led to study the logarithmic norm of a set of products of type $D A$, where $D$ is a diagonal matrix such that ${\mathrm{diag}}(D) \in σ'({\mathbb R}^n)$. Specifically, given a real number $c$ (usually $c=0$), the problem consists in finding the largest positive interval $\text{I}\subseteq \mathbb [0,\infty)$ such that the logarithmic norm $μ(DA) \le c$ for all diagonal matrices $D$ with $D_{ii}\in \text{I}$. We propose a two-level nested methodology: an inner level where, for a given $\text{I}$, we compute an optimizer $D^\star(\text{I})$ by a gradient system approach, and an outer level where we tune $\text{I}$ so that the value $c$ is reached by $μ(D^\star(\text{I})A)$. We extend the proposed two-level approach to the general multilayer, and possibly time-dependent, case $\dot{u}(t) = σ( A_k(t) \ldots σ( A_{1}(t) u(t) + b_{1}(t) ) \ldots + b_{k}(t) )$ and we propose several numerical examples to illustrate its behaviour, including its stabilizing performance on a one-layer neural ODE applied to the classification of the MNIST handwritten digits dataset.

math.NA↗
arXiv2024

Cholesky-like Preconditioner for Hodge Laplacians via Heavy Collapsible Subcomplex

Techniques based on $k$-th order Hodge Laplacian operators $L_k$ are widely used to describe the topology as well as the governing dynamics of high-order systems modeled as simplicial complexes. In all of them, it is required to solve a number of least square problems with $L_k$ as coefficient matrix, for example in order to compute some portions of the spectrum or integrate the dynamical system. In this work, we introduce the notion of optimal collapsible subcomplex and we present a fast combinatorial algorithm for the computation of a sparse Cholesky-like preconditioner for $L_k$ that exploits the topological structure of the simplicial complex. The performance of the preconditioner is tested for conjugate gradient method for least square problems (CGLS) on a variety of simplicial complexes with different dimensions and edge densities. We show that, for sparse simplicial complexes, the new preconditioner reduces significantly the condition number of $L_k$ and performs better than the standard incomplete Cholesky factorization.

math.NA↗
arXiv2023

Quantifying the structural stability of simplicial homology

The homology groups of a simplicial complex reveal fundamental properties of the topology of the data or the system and the notion of topological stability naturally poses an important yet not fully investigated question. In the current work, we study the stability in terms of the smallest perturbation sufficient to change the dimensionality of the corresponding homology group. Such definition requires an appropriate weighting and normalizing procedure for the boundary operators acting on the Hodge algebra's homology groups. Using the resulting boundary operators, we then formulate the question of structural stability as a spectral matrix nearness problem for the corresponding higher-order graph Laplacian. We develop a bilevel optimization procedure suitable for the formulated matrix nearness problem and illustrate the method's performance on a variety of synthetic quasi-triangulation datasets and transportation networks.

math.NA↗
arXiv2020

Infinite energy solutions for weakly damped quintic wave equations in $\mathbb{R}^3$

The paper gives a comprehensive study of infinite-energy solutions and their long-time behavior for semi-linear weakly damped wave equations in $\mathbb{R}^3$ with quintic nonlinearities. This study includes global well-posedness of the so-called Shatah-Struwe solutions, their dissipativity, the existence of a locally compact global attractors (in the uniformly local phase spaces) and their extra regularity.

math.AP↗
arXiv2018

Homogenisation with error estimates of attractors for damped semi-linear anisotropic wave equations

Homogenisation of global $\mathcal{A}^ε$ and exponential $\mathcal{M}^ε$ attractors for the damped semi-linear anisotropic wave equation $\partial_t^2 u^ε+γ\partial_t u^ε-{\rm div} \left(a\left( \tfrac{x}ε \right)\nabla u^ε\right)+f(u^ε)=g$, on a bounded domain $Ω\subset \mathbb{R}^3$, is performed. Order-sharp estimates between trajectories $u^ε(t)$ and their homogenised trajectories $u^0(t)$ are established. These estimates are given in terms of the operator-norm difference between resolvents of the elliptic operator ${\rm div}\left(a\left( \tfrac{x}ε \right)\nabla \right)$ and its homogenised limit ${\rm div}\left(a^h\nabla \right)$. Consequently, norm-resolvent estimates on the Hausdorff distance between the anisotropic attractors and their homogenised counter-parts $\mathcal{A}^0$ and $\mathcal{M}^0$ are established. These results imply error estimates of the form ${\rm dist}_X(\mathcal{A}^ε, \mathcal{A}^0) \le C ε^\varkappa$ and ${\rm dist}^s_X(\mathcal{M}^ε, \mathcal{M}^0) \le C ε^\varkappa$ in the spaces $X =L^2(Ω)\times H^{-1}(Ω)$ and $X =(C^β(\overlineΩ))^2$. In the natural energy space $\mathcal{E} : = H^1_0(Ω) \times L^2(Ω)$, error estimates ${\rm dist}_{\mathcal{E}}(\mathcal{A}^ε, {T}_ε\mathcal{A}^0) \le C \sqrtε^\varkappa$ and ${\rm dist}^s_{\mathcal{E}}(\mathcal{M}^ε, {T}_ε\mathcal{M}^0) \le C \sqrtε^\varkappa$ are established where ${T}_ε$ is first-order correction for the homogenised attractors suggested by asymptotic expansions. Our results are applied to Dirchlet, Neumann and periodic boundary conditions.

math.AP↗
arXiv2015

Infinite energy solutions for critical wave equation with fractional damping in unbounded domains

This work is devoted to infinite-energy solutions of semi-linear wave equations in unbounded smooth domains of $\mathbb{R}^3$ with fractional damping of the form $(-Δ_x+1)^\frac{1}{2}\partial_t u$. The work extends previously known results for bounded domains in finite energy case. Furthermore, well-posedness and existence of locally-compact smooth attractors for the critical quintic non-linearity are obtained under less restrictive assumptions on non-linearity, relaxing some artificial technical conditions used before. This is achieved by virtue of new type Lyapunov functional that allows to establish extra space-time regularity of solutions of Strichartz type.

math.AP↗
arXiv2015

Finite Dimensionality of the attractor for the hyperbolic Cahn-Hilliard-Oono Equation in R^3

In this paper, we continue the study of the hyperbolic relaxation of the Cahn-Hilliard-Oono equation with the sub-quintic non-linearity in the whole space $\R^3$ started in our previous paper and verify that under the natural assumptions on the non-linearity and the external force, the fractal dimension of the associated global attractor in the natural energy space is finite.

math.AP↗
arXiv2014

Global well-posedness and attractors for the hyperbolic Cahn-Hilliard-Oono equation in the whole space

We prove the global well-posedness of the so-called hyperbolic relaxation of the Cahn-Hilliard-Oono equation in the whole space R^3 with the non-linearity of the sub-quintic growth rate. Moreover, the dissipativity and the existence of a smooth global attractor in the naturally defined energy space is also verified. The result is crucially based on the Strichartz estimates for the linear Scroedinger equation in R^3.

math.AP↗
arXiv2014

Strichartz estimates and smooth attractors for a sub-quintic wave equation with fractional damping in bounded domains

The work is devoted to Dirichlet problem for sub-quintic semi-linear wave equation with damping damping term of the form $(-Δ)^α\partial_t u$, $α\in(0,\frac{1}{2})$, in bounded smooth domains of $\Bbb R^3$. It appears that to prove well-posedness and develop smooth attractor theory for the problem we need additional regularity of the solutions, which does not follow from the energy estimate. Considering the original problem as perturbation of the linear one the task is reduced to derivation of Strichartz type estimate for the linear wave equation with fractional damping, which is the main feature of the work. Existence of smooth exponential attractor for the natural dynamical system associated with the problem is also established.

math.AP↗
arXiv2013

Attractors for damped quintic wave equations in bounded domains

The dissipative wave equation with a critical quintic nonlinearity in smooth bounded three dimensional domain is considered. Based on the recent extension of the Strichartz estimates to the case of bounded domains, the existence of a compact global attractor for the solution semigroup of this equation is established. Moreover, the smoothness of the obtained attractor is also shown.

math.AP↗
arXiv2013

Smooth attractors for the quintic wave equations with fractional damping

Dissipative wave equations with critical quintic nonlinearity and damping term involving the fractional Laplacian are considered. The additional regularity of energy solutions is established by constructing the new Lyapunov-type functional and based on this, the global well-posedness and dissipativity of the energy solutions as well as the existence of a smooth global and exponential attractors of finite Hausdorff and fractal dimension is verified.

math.AP↗