Search

We propose an algorithmic framework, that employs active subspace techniques, for scalable global optimization of functions with low effective dimension (also referred to as low-rank functions). This proposal replaces the original high-dimensional problem by one or several lower-dimensional reduced subproblem(s), capturing the main directions of variation of the objective which are estimated here as the principal components of a collection of sampled gradients. We quantify the sampling complexity of estimating the subspace of variation of the objective in terms of its effective dimension and hence, bound the probability that the reduced problem will provide a solution to the original problem. To account for the practical case when the effective dimension is not known a priori, our framework adaptively solves a succession of reduced problems, increasing the number of sampled gradients until the estimated subspace of variation remains unchanged. We prove global convergence under mild assumptions on the objective, the sampling distribution and the subproblem solver, and illustrate numerically the benefits of our proposed algorithms over those using random embeddings.

In fibrous paper materials, an exposure to a variation in moisture content causes changes in the geometrical and mechanical properties. Such changes are strongly affected by the inter-fibre bonds, which are responsible for the transfer of the hygro-mechanical response from one fibre to its neighbours in the network, resulting in sheet-scale deformation. Most models developed in literature assume perfect bonding between fibres. In the 3D reality, there is some flexibility in the bond region, even for the perfectly bonded fibres, because of the possibility of deformation gradients through the fibre thickness. In earlier 2D idealizations, perfectly bonded fibres were assumed, implying full kinematic constraint through the entire thickness of the sheet. The purpose of the present study is to assess the effect of this assumption. Using a homogenization approach, a random network of fibres is generated with different coverages and modelled using finite elements. In order to understand the role of bonding between fibres on the hygro-expansive behaviour of a network, a bond model is developed. In this model, the fibres are modelled using 2D regular bulk finite elements and the bonds are represented by interfacial elements of finite stiffness, which are introduced between each pair of fibres bonded in the network. These embedded interfacial elements form a connection between two respective fibres, allowing relative displacements between their mid-planes. The hygro-elastic response of networks obtained with this bond model is investigated by varying the bond stiffness and the network coverage under the application of mechanical loading and changes in moisture content. Furthermore, the bond model is used to analyse the influence of inter-fibre bonds on the anisotropic response of the paper fibre network., Comment: 26 pages, 16 figures

Positive-unlabeled learning (PU learning) is known as a special case of semi-supervised binary classification where only a fraction of positive examples are labeled. The challenge is then to find the correct classifier despite this lack of information. Recently, new methodologies have been introduced to address the case where the probability of being labeled may depend on the covariates. In this paper, we are interested in establishing risk bounds for PU learning under this general assumption. In addition, we quantify the impact of label noise on PU learning compared to standard classification setting. Finally, we provide a lower bound on minimax risk proving that the upper bound is almost optimal.

We study the function $$\mbox{KVol} : (X,\omega)\mapsto \mbox{Vol} (X,\omega) \sup_{\alpha,\beta} \frac{\mbox{Int} (\alpha,\beta)}{l_g (\alpha) l_g (\beta)}$$ defined on the moduli spaces of translation surfaces. More precisely, let $\mathcal T_n$ be the Teichm\"uller discs of the original Veech surface $(X_n,\omega_n)$ arising from right-angled triangle with angles $(\pi/2,\pi/n,(n-2)\pi/2n)$ by the unfolding construction for $n\geq 5$. For $n \equiv 1 \mod 2$ and any $(X,\omega)\in \mathcal T_n$, we establish the (sharp) bounds $$ \frac{n}{2} \cot \frac{\pi}{n} \leq \mbox{KVol}(X,\omega) \leq \frac{n}{2} \cot \frac{\pi}{n} \cdot \frac1{\sin \frac{2\pi}{n}}.$$ The lower bound is uniquely realized at $(X_n,\omega_n)$., Comment: New version with the first author added and completely different methods. We focus on the $n=2m+1$ case, the $n=4m$ case is dealt with in a forthcoming paper by the first author. 30pages, 15 figures

The study of the gravitational field produced by a spatially non-local, superposed quantum state of a massive particle is a thrilling area of modern physics. One question to be answered is whether the gravitational field behaves as the classical superposition of two particles separated by a spatial distance with half the mass located at each position or as a quantum superposition with a far more interesting and subtle behaviour for the gravitational field. Quantum field theory is ideally suited to probe exactly this kind of question. We study the scattering of a massless scalar on such a spatially nonlocal, quantum superposition of a massive particle. We compute the differential scattering cross section corresponding to the interaction coming from the exchange of one graviton. We find that the scattering cross section is not at all represented by the Schr\"odinger-Newton picture of potential scattering from two localized sources with half the mass at each source. We discuss how our result would be lethal to the Schr\"odinger-Newton description of gravitation interacting with quantum matter and would be conducive to considering the gravitational field to be quantized. We comment on the experimental feasibility of observing such effects., Comment: 6 pages, 1 figure

We propose to use stochastic Riemannian coordinate descent on the orthogonal group for recurrent neural network training. The algorithm rotates successively two columns of the recurrent matrix, an operation that can be efficiently implemented as a multiplication by a Givens matrix. In the case when the coordinate is selected uniformly at random at each iteration, we prove the convergence of the proposed algorithm under standard assumptions on the loss function, stepsize and minibatch noise. In addition, we numerically demonstrate that the Riemannian gradient in recurrent neural network training has an approximately sparse structure. Leveraging this observation, we propose a faster variant of the proposed algorithm that relies on the Gauss-Southwell rule. Experiments on a benchmark recurrent neural network training problem are presented to demonstrate the effectiveness of the proposed algorithm.

We propose a random-subspace algorithmic framework for global optimization of Lipschitz-continuous objectives, and analyse its convergence using novel tools from conic integral geometry. X-REGO randomly projects, in a sequential or simultaneous manner, the high-dimensional original problem into low-dimensional subproblems that can then be solved with any global, or even local, optimization solver. We estimate the probability that the randomly-embedded subproblem shares (approximately) the same global optimum as the original problem. This success probability is then used to show convergence of X-REGO to an approximate global solution of the original problem, under weak assumptions on the problem (having a strictly feasible global solution) and on the solver (guaranteed to find an approximate global solution of the reduced problem with sufficiently high probability). In the particular case of unconstrained objectives with low effective dimension, that only vary over a low-dimensional subspace, we propose an X-REGO variant that explores random subspaces of increasing dimension until finding the effective dimension of the problem, leading to X-REGO globally converging after a finite number of embeddings, proportional to the effective dimension. We show numerically that this variant efficiently finds both the effective dimension and an approximate global minimizer of the original problem., Comment: 41 pages

We review the different notions about translation surfaces which are necessary to understand McMullen's classification of $GL_2^+(\mathbb{R})$-orbit closures in genus two. In Section 2 we recall the different definitions of a translation surface, in increasing order of abstraction, starting with cutting and pasting plane polygons, ending with Abelian differentials. In Section 3 we define the moduli space of translation surfaces and explain its stratification by the type of zeroes of the Abelian differential, the local coordinates given by the relative periods, its relationship with the moduli space of complex structures and the Teichm\H{u}ller geodesic flow. In Part II we introduce the $GL_2^+(\mathbb{R})$-action, and define the related notions of Veech group, Teichm\H{u}ller disk, and Veech surface. In Section 9 we explain how McMullen classifies $GL_2^+(\mathbb{R})$-orbit closures in genus $2$: you have orbit closures of dimension $1$ (Veech surfaces, of which a complete list is given), $2$ (Hilbert modular surfaces, of which again a complete list is given), and $3$ (the whole moduli space of complex structures). In the last section we review some recent progress in higher genus., Comment: The final version of this paper will appear in the book: Surveys in Geometry I, ed. A. Papadopoulos, Springer Nature, 2021

The use of Internet of Things (IoT) technologies is becoming a preferred solution for the assessment of tailings dams' safety. Real-time sensor monitoring proves to be a key tool for reducing the risk related to these ever-evolving earth-fill structures, that exhibit a high rate of sudden and hazardous failures. In order to optimally exploit real-time embankment monitoring, one major hindrance has to be overcome: the creation of a supporting numerical model for stability analysis, with rapid-enough response to perform data assimilation in real time. A model should be built, such that its response can be obtained faster than the physical evolution of the analyzed phenomenon. In this work, Reduced Order Modelling (ROM) is used to boost computational efficiency in solving the coupled hydro-mechanical system of equations governing the problem. The Reduced Basis method is applied to the coupled hydro-mechanical equations that govern the groundwater flow, that are made non-linear as a result of considering an unsaturated soil. The resulting model's performance is assessed by solving a 2D and a 3D problem relevant to tailings dams' safety. The ROM technique achieves a speedup of 3 to 15 times with respect to the full-order model (FOM) while maintaining high levels of accuracy.

This paper considers a multi-patch model, where each patch follows a logistic law, and patches are coupled by asymmetrical migration terms. First, in the case of perfect mixing, i.e when the migration rate tends to infinity, the total population follows a logistic equation with a carrying capacity which in general is different from the sum of the n carrying capacities, and depends on the migration terms. Second, we determine, in some particular cases, the conditions under which fragmentation and asymmetrical migration can lead to a total equilibrium population greater or smaller than the sum of the carrying capacities. Finally, for the three-patch model, we show numerically the existence of at least three critical values of the migration rate for which the total equilibrium population equals the sum of the carrying capacities., Comment: 29 pages, 7 figures, 1 table. Section 5 added in version 2

Reversibility is of paramount importance in the correct representation of surface peeling in various physical settings, ranging from motility in nature, to gripping devices in robotic applications, and even to sliding of tectonic plates. Modeling the detachment-reattachment sequence, known as stick-slip, imposes several challenges in a continuum framework. Here we exploit customized reversible cohesive elements in a hybrid finite element model that can handle occurrence of snap-through instabilities. The simulations capture various peeling phenomena that emerge in experimental observations, where layers are pulled from a flat, rigid substrate in the direction parallel to the surface. For long layers, periodicity in reattachment is shown to develop and is linked to the concept of Schallamach waves. Further, the connection between surface properties and stick-slip behavior is investigated: we find that stick-slip is linked to the propensity of the interface to localize deformation and damage. Beyond elucidating the various peeling behaviors and the detachment modes, the computational framework developed here provides a straightforward approach for investigation of complex delamination processes, which can guide development of future applications across different scales and in various settings.

A dynamical resolution to the cosmological constant fine-tuning problem has been previously put forward, based on a scalar-tensor gravitational theory possessing de Sitter attractor solutions characterized by a small Hubble expansion rate, irrespective of an initially large vacuum energy. We show that a technically natural subregion of the parameter space yields a cosmological evolution through radiation- and matter-dominated eras that is essentially indistinguishable from that predicted by General Relativity. Similarly, the proposed model automatically satisfies the observational constraints on a fifth force mediated by the new scalar degree of freedom., Comment: 15 pages, 2 figures; v2: extended discussions, matches the published version

We consider the bound-constrained global optimization of functions with low effective dimensionality, that are constant along an (unknown) linear subspace and only vary over the effective (complement) subspace. We aim to implicitly explore the intrinsic low dimensionality of the constrained landscape using feasible random embeddings, in order to understand and improve the scalability of algorithms for the global optimization of these special-structure problems. A reduced subproblem formulation is investigated that solves the original problem over a random low-dimensional subspace subject to affine constraints, so as to preserve feasibility with respect to the given domain. Under reasonable assumptions, we show that the probability that the reduced problem is successful in solving the original, full-dimensional problem is positive. Furthermore, in the case when the objective's effective subspace is aligned with the coordinate axes, we provide an asymptotic bound on this success probability that captures its algebraic dependence on the effective and, surprisingly, ambient dimensions. We then propose X-REGO, a generic algorithmic framework that uses multiple random embeddings, solving the above reduced problem repeatedly, approximately and possibly, adaptively. Using the success probability of the reduced subproblems, we prove that X-REGO converges globally, with probability one, and linearly in the number of embeddings, to an $\epsilon$-neighbourhood of a constrained global minimizer. Our numerical experiments on special structure functions illustrate our theoretical findings and the improved scalability of X-REGO variants when coupled with state-of-the-art global - and even local - optimization solvers for the subproblems.

We study the quantity $\mbox{KVol}$ defined as the supremum, over all pairs of closed curves, of their algebraic intersection, divided by the product of their lengths, times the area of the surface. The surfaces we consider live in the stratum $\mathcal{H}(2)$ of translation surfaces of genus $2$, with one conical point. We provide an explicit sequence $L(n,n)$ of surfaces such that $\mbox{KVol}(L(n,n)) \longrightarrow 2$ when $n$ goes to infinity, $2$ being the conjectured infimum for $\mbox{KVol}$ over $\mathcal{H}(2)$., Comment: 8 pages, 3 figures. arXiv admin note: text overlap with arXiv:2007.10847

The setting is a square-tiled surface X. We study the quantity KVol, defined as the supremum over all pairs of closed curves, of their algebraic intersection divided by the product of their length, times the volume of X (so as to make it scaling-invariant). We give a hyperbolic-geometric construction to compute KVol in a family of Teichm\H{u}ller disks of square-tiled surfaces., Comment: 25 pages, 15 figures

Materials like paper, consisting of a network of natural fibres, exposed to variations in moisture, undergo changes in geometrical and mechanical properties. This behaviour is particularly important for understanding the hygro-mechanical response of sheets of paper in applications like digital printing. A two-dimensional microstructural model of a fibrous network is therefore developed to upscale the hygro-expansion of individual fibres, through their interaction, to the resulting overall expansion of the network. The fibres are modelled with rectangular shapes and are assumed to be perfectly bonded where they overlap. For realistic networks the number of bonds is large and the network is geometrically so complex that discretizing it by conventional, geometry-conforming, finite elements is cumbersome. The combination of a level-set and XFEM formalism enables the use of regular, structured grids in order to model the complex microstructural geometry. In this approach, the fibres are described implicitly by a level-set function. In order to represent the fibre boundaries in the fibrous network, an XFEM discretization is used together with a Heaviside enrichment function. Numerical results demonstrate that the proposed approach successfully captures the hygro-expansive properties of the network with fewer degrees of freedom compared to classical FEM, preserving desired accuracy., Comment: 27 pages, 22 figures, 4 tables, J. Appl. Mech. June 19, 2020

We propose a differential geometric construction for families of low-rank covariance matrices, via interpolation on low-rank matrix manifolds. In contrast with standard parametric covariance classes, these families offer significant flexibility for problem-specific tailoring via the choice of "anchor" matrices for the interpolation. Moreover, their low-rank facilitates computational tractability in high dimensions and with limited data. We employ these covariance families for both interpolation and identification, where the latter problem comprises selecting the most representative member of the covariance family given a data set. In this setting, standard procedures such as maximum likelihood estimation are nontrivial because the covariance family is rank-deficient; we resolve this issue by casting the identification problem as distance minimization. We demonstrate the power of these differential geometric families for interpolation and identification in a practical application: wind field covariance approximation for unmanned aerial vehicle navigation.

When balanced truncation is used for model order reduction, one has to solve a pair of Lyapunov equations for two Gramians and uses them to construct a reduced-order model. Although advances in solving such equations have been made, it is still the most expensive step of this reduction method. Parametric model order reduction aims to determine reduced-order models for parameter-dependent systems. Popular techniques for parametric model order reduction rely on interpolation. Nevertheless, the interpolation of Gramians is rarely mentioned, most probably due to the fact that Gramians are symmetric positive semidefinite matrices, a property that should be preserved by the interpolation method. In this contribution, we propose and compare two approaches for Gramian interpolation. In the first approach, the interpolated Gramian is computed as a linear combination of the data Gramians with positive coefficients. Even though positive semidefiniteness is guaranteed in this method, the rank of the interpolated Gramian can be significantly larger than that of the data Gramians. The second approach aims to tackle this issue by performing the interpolation on the manifold of fixed-rank positive semidefinite matrices. The results of the interpolation step are then used to construct parametric reduced-order models, which are compared numerically on two benchmark problems., Comment: 19 pages, 7 figures, submitted as a book chapter