arXiv subjects

Afonso S. Bandeira

Publications and source records attributed to Afonso S. Bandeira.

At least 55 records · Page 3Linked to original sources

arXiv2020

Average-Case Integrality Gap for Non-Negative Principal Component Analysis

Montanari and Richard (2015) asked whether a natural semidefinite programming (SDP) relaxation can effectively optimize $\mathbf{x}^{\top}\mathbf{W} \mathbf{x}$ over $\|\mathbf{x}\| = 1$ with $x_i \geq 0$ for all coordinates $i$, where $\mathbf{W} \in \mathbb{R}^{n \times n}$ is drawn from the Gaussian orthogonal ensemble (GOE) or a spiked matrix model. In small numerical experiments, this SDP appears to be tight for the GOE, producing a rank-one optimal matrix solution aligned with the optimal vector $\mathbf{x}$. We prove, however, that as $n \to \infty$ the SDP is not tight, and certifies an upper bound asymptotically no better than the simple spectral bound $λ_{\max}(\mathbf{W})$ on this objective function. We also provide evidence, using tools from recent literature on hypothesis testing with low-degree polynomials, that no subexponential-time certification algorithm can improve on this behavior. Finally, we present further numerical experiments estimating how large $n$ would need to be before this limiting behavior becomes evident, providing a cautionary example against extrapolating asymptotics of SDPs in high dimension from their efficacy in small "laptop scale" computations.

cs.DS↗

arXiv2020

A Tight Degree 4 Sum-of-Squares Lower Bound for the Sherrington-Kirkpatrick Hamiltonian

We show that, if $\mathbf{W} \in \mathbb{R}^{N \times N}_{\mathsf{sym}}$ is drawn from the gaussian orthogonal ensemble, then with high probability the degree 4 sum-of-squares relaxation cannot certify an upper bound on the objective $N^{-1} \cdot \mathbf{x}^\top \mathbf{W} \mathbf{x}$ under the constraints $x_i^2 - 1 = 0$ (i.e. $\mathbf{x} \in \{ \pm 1 \}^N$) that is asymptotically smaller than $λ_{\max}(\mathbf{W}) \approx 2$. We also conjecture a proof technique for lower bounds against sum-of-squares relaxations of any degree held constant as $N \to \infty$, by proposing an approximate pseudomoment construction.

cs.DS↗

arXiv2020

Spectral Planting and the Hardness of Refuting Cuts, Colorability, and Communities in Random Graphs

We study the problem of efficiently refuting the k-colorability of a graph, or equivalently certifying a lower bound on its chromatic number. We give formal evidence of average-case computational hardness for this problem in sparse random regular graphs, showing optimality of a simple spectral certificate. This evidence takes the form of a computationally-quiet planting: we construct a distribution of d-regular graphs that has significantly smaller chromatic number than a typical regular graph drawn uniformly at random, while providing evidence that these two distributions are indistinguishable by a large class of algorithms. We generalize our results to the more general problem of certifying an upper bound on the maximum k-cut. This quiet planting is achieved by minimizing the effect of the planted structure (e.g. colorings or cuts) on the graph spectrum. Specifically, the planted structure corresponds exactly to eigenvectors of the adjacency matrix. This avoids the pushout effect of random matrix theory, and delays the point at which the planting becomes visible in the spectrum or local statistics. To illustrate this further, we give similar results for a Gaussian analogue of this problem: a quiet version of the spiked model, where we plant an eigenspace rather than adding a generic low-rank perturbation. Our evidence for computational hardness of distinguishing two distributions is based on three different heuristics: stability of belief propagation, the local statistics hierarchy, and the low-degree likelihood ratio. Of independent interest, our results include general-purpose bounds on the low-degree likelihood ratio for multi-spiked matrix models, and an improved low-degree analysis of the stochastic block model.

cs.CC↗

arXiv2020

Spurious Valleys in Two-layer Neural Network Optimization Landscapes

Neural networks provide a rich class of high-dimensional, non-convex optimization problems. Despite their non-convexity, gradient-descent methods often successfully optimize these models. This has motivated a recent spur in research attempting to characterize properties of their loss surface that may explain such success. In this paper, we address this phenomenon by studying a key topological property of the loss: the presence or absence of spurious valleys, defined as connected components of sub-level sets that do not include a global minimum. Focusing on a class of two-layer neural networks defined by smooth (but generally non-linear) activation functions, we identify a notion of intrinsic dimension and show that it provides necessary and sufficient conditions for the absence of spurious valleys. More concretely, finite intrinsic dimension guarantees that for sufficiently overparametrised models no spurious valleys exist, independently of the data distribution. Conversely, infinite intrinsic dimension implies that spurious valleys do exist for certain data distributions, independently of model overparametrisation. Besides these positive and negative results, we show that, although spurious valleys may exist in general, they are confined to low risk levels and avoided with high probability on overparametrised models.

math.OC↗

arXiv2019

Experimental performance of graph neural networks on random instances of max-cut

This note explores the applicability of unsupervised machine learning techniques towards hard optimization problems on random inputs. In particular we consider Graph Neural Networks (GNNs) -- a class of neural networks designed to learn functions on graphs -- and we apply them to the max-cut problem on random regular graphs. We focus on the max-cut problem on random regular graphs because it is a fundamental problem that has been widely studied. In particular, even though there is no known explicit solution to compare the output of our algorithm to, we can leverage the known asymptotics of the optimal max-cut value in order to evaluate the performance of the GNNs. In order to put the performance of the GNNs in context, we compare it with the classical semidefinite relaxation approach by Goemans and Williamson~(SDP), and with extremal optimization, which is a local optimization heuristic from the statistical physics literature. The numerical results we obtain indicate that, surprisingly, Graph Neural Networks attain comparable performance to the Goemans and Williamson SDP. We also observe that extremal optimization consistently outperforms the other two methods. Furthermore, the performances of the three methods present similar patterns, that is, for sparser, and for larger graphs, the size of the found cuts are closer to the asymptotic optimal max-cut value.

math.OC↗

arXiv2019

Notes on Computational Hardness of Hypothesis Testing: Predictions using the Low-Degree Likelihood Ratio

These notes survey and explore an emerging method, which we call the low-degree method, for predicting and understanding statistical-versus-computational tradeoffs in high-dimensional inference problems. In short, the method posits that a certain quantity -- the second moment of the low-degree likelihood ratio -- gives insight into how much computational time is required to solve a given hypothesis testing problem, which can in turn be used to predict the computational hardness of a variety of statistical inference tasks. While this method originated in the study of the sum-of-squares (SoS) hierarchy of convex programs, we present a self-contained introduction that does not require knowledge of SoS. In addition to showing how to carry out predictions using the method, we include a discussion investigating both rigorous and conjectural consequences of these predictions. These notes include some new results, simplified proofs, and refined conjectures. For instance, we point out a formal connection between spectral methods and the low-degree likelihood ratio, and we give a sharp low-degree lower bound against subexponential-time algorithms for tensor PCA.

math.ST↗

arXiv2019

The sample complexity of multi-reference alignment

The growing role of data-driven approaches to scientific discovery has unveiled a large class of models that involve latent transformations with a rigid algebraic constraint. Three-dimensional molecule reconstruction in Cryo-Electron Microscopy (cryo-EM) is a central problem in this class. Despite decades of algorithmic and software development, there is still little theoretical understanding of the sample complexity of this problem, that is, number of images required for 3-D reconstruction. Here we consider multi-reference alignment (MRA), a simple model that captures fundamental aspects of the statistical and algorithmic challenges arising in cryo-EM and related problems. In MRA, an unknown signal is subject to two types of corruption: a latent cyclic shift and the more traditional additive white noise. The goal is to recover the signal at a certain precision from independent samples. While at high signal-to-noise ratio (SNR), the number of observations needed to recover a generic signal is proportional to $1/\mathrm{SNR}$, we prove that it rises to a surprising $1/\mathrm{SNR}^3$ in the low SNR regime. This precise phenomenon was observed empirically more than twenty years ago for cryo-EM but has remained unexplained to date. Furthermore, our techniques can easily be extended to the heterogeneous MRA model where the samples come from a mixture of signals, as is often the case in applications such as cryo-EM, where molecules may have different conformations. This provides a first step towards a statistical theory for heterogeneous cryo-EM.

cs.IT↗

arXiv2019

Deterministic guarantees for Burer-Monteiro factorizations of smooth semidefinite programs

We consider semidefinite programs (SDPs) with equality constraints. The variable to be optimized is a positive semidefinite matrix $X$ of size $n$. Following the Burer--Monteiro approach, we optimize a factor $Y$ of size $n \times p$ instead, such that $X = YY^T$. This ensures positive semidefiniteness at no cost and can reduce the dimension of the problem if $p$ is small, but results in a non-convex optimization problem with a quadratic cost function and quadratic equality constraints in $Y$. In this paper, we show that if the set of constraints on $Y$ regularly defines a smooth manifold, then, despite non-convexity, first- and second-order necessary optimality conditions are also sufficient, provided $p$ is large enough. For smaller values of $p$, we show a similar result holds for almost all (linear) cost functions. Under those conditions, a global optimum $Y$ maps to a global optimum $X = YY^T$ of the SDP. We deduce old and new consequences for SDP relaxations of the generalized eigenvector problem, the trust-region subproblem and quadratic optimization over several spheres, as well as for the Max-Cut and Orthogonal-Cut SDPs which are common relaxations in stochastic block modeling and synchronization of rotations.

math.OC↗

arXiv2019

On the Landscape of Synchronization Networks: A Perspective from Nonconvex Optimization

Studying the landscape of nonconvex cost function is key towards a better understanding of optimization algorithms widely used in signal processing, statistics, and machine learning. Meanwhile, the famous Kuramoto model has been an important mathematical model to study the synchronization phenomena of coupled oscillators over various network topologies. In this paper, we bring together these two seemingly unrelated objects by investigating the optimization landscape of a nonlinear function $E(\boldsymbolθ) = \frac{1}{2}\sum_{1\leq i,j\leq n} a_{ij}(1-\cos(θ_i - θ_j))$ associated to an underlying network and exploring the relationship between the existence of local minima and network topology. This function arises naturally in Burer-Monteiro method applied to $\mathbb{Z}_2$ synchronization as well as matrix completion on the torus. Moreover, it corresponds to the energy function of the homogeneous Kuramoto model on complex networks for coupled oscillators. We prove the minimizer of the energy function is unique up to a global translation under deterministic dense graphs and Erdős-Rényi random graphs with tools from optimization and random matrix theory. Consequently, the stable equilibrium of the corresponding homogeneous Kuramoto model is unique and the basin of attraction for the synchronous state of these coupled oscillators is the whole phase space minus a set of measure zero. In addition, our results address when the Burer-Monteiro method recovers the ground truth exactly from highly incomplete observations in $\mathbb{Z}_2$ synchronization and shed light on the robustness of nonconvex optimization algorithms against certain types of so-called monotone adversaries. Numerical simulations are performed to illustrate our results.

math.OC↗

arXiv2019

Computational Hardness of Certifying Bounds on Constrained PCA Problems

Given a random $n \times n$ symmetric matrix $\boldsymbol W$ drawn from the Gaussian orthogonal ensemble (GOE), we consider the problem of certifying an upper bound on the maximum value of the quadratic form $\boldsymbol x^\top \boldsymbol W \boldsymbol x$ over all vectors $\boldsymbol x$ in a constraint set $\mathcal{S} \subset \mathbb{R}^n$. For a certain class of normalized constraint sets $\mathcal{S}$ we show that, conditional on certain complexity-theoretic assumptions, there is no polynomial-time algorithm certifying a better upper bound than the largest eigenvalue of $\boldsymbol W$. A notable special case included in our results is the hypercube $\mathcal{S} = \{ \pm 1 / \sqrt{n}\}^n$, which corresponds to the problem of certifying bounds on the Hamiltonian of the Sherrington-Kirkpatrick spin glass model from statistical physics. Our proof proceeds in two steps. First, we give a reduction from the detection problem in the negatively-spiked Wishart model to the above certification problem. We then give evidence that this Wishart detection problem is computationally hard below the classical spectral threshold, by showing that no low-degree polynomial can (in expectation) distinguish the spiked and unspiked models. This method for identifying computational thresholds was proposed in a sequence of recent works on the sum-of-squares hierarchy, and is believed to be correct for a large class of problems. Our proof can be seen as constructing a distribution over symmetric matrices that appears computationally indistinguishable from the GOE, yet is supported on matrices whose maximum quadratic form over $\boldsymbol x \in \mathcal{S}$ is much larger than that of a GOE matrix.

cs.DS↗

arXiv2019

A Gramian Description of the Degree 4 Generalized Elliptope

One of the most widely studied convex relaxations in combinatorial optimization is the relaxation of the cut polytope $\mathscr C^N$ to the elliptope $\mathscr E^N$, which corresponds to the degree 2 sum-of-squares (SOS) relaxation of optimizing a quadratic form over the hypercube $\{\pm 1\}^N$. We study the extension of this classical idea to degree 4 SOS, which gives an intermediate relaxation we call the degree 4 generalized elliptope $\mathscr E_4^N$. Our main result is a necessary and sufficient condition for the Gram matrix of a collection of vectors to belong to $\mathscr E_4^N$. Consequences include a tight rank inequality between degree 2 and degree 4 pseudomoment matrices, and a guarantee that the only extreme points of $\mathscr E^N$ also in $\mathscr E_4^N$ are the cut matrices; that is, $\mathscr E^N$ and $\mathscr E_4^N$ share no "spurious" extreme point. For Gram matrices of equiangular tight frames, we give a simple criterion for membership in $\mathscr{E}_4^N$. This yields new inequalities satisfied in $\mathscr{E}_4^N$ but not $\mathscr{E}^N$ whose structure is related to the Schläfli graph and which cannot be obtained as linear combinations of triangle inequalities. We also give a new proof of the restriction to degree 4 of a result of Laurent showing that $\mathscr{E}_4^N$ does not satisfy certain cut polytope inequalities capturing parity constraints. Though limited to this special case, our proof of the positive semidefiniteness of Laurent's pseudomoment matrix is short and elementary. Our techniques also suggest that membership in $\mathscr{E}_4^N$ is closely related to the partial transpose operation on block matrices, which has previously played an important role in the study of quantum entanglement. To illustrate, we present a correspondence between certain entangled bipartite quantum states and the matrices of $\mathscr{E}_4^N\setminus\mathscr{C}^N$.

math.OC↗

arXiv2019

Sum-of-Squares Optimization and the Sparsity Structure of Equiangular Tight Frames

Equiangular tight frames (ETFs) may be used to construct examples of feasible points for semidefinite programs arising in sum-of-squares (SOS) optimization. We show how generalizing the calculations in a recent work of the authors' that explored this connection also yields new bounds on the sparsity of (both real and complex) ETFs. One corollary shows that Steiner ETFs corresponding to finite projective planes are optimally sparse in the sense of achieving tightness in a matrix inequality controlling overlaps between sparsity patterns of distinct rows of the synthesis matrix. We also formulate several natural open problems concerning further generalizations of our technique.

math.FA↗

arXiv2018

Revised Note on Learning Algorithms for Quadratic Assignment with Graph Neural Networks

Inverse problems correspond to a certain type of optimization problems formulated over appropriate input distributions. Recently, there has been a growing interest in understanding the computational hardness of these optimization problems, not only in the worst case, but in an average-complexity sense under this same input distribution. In this revised note, we are interested in studying another aspect of hardness, related to the ability to learn how to solve a problem by simply observing a collection of previously solved instances. These 'planted solutions' are used to supervise the training of an appropriate predictive model that parametrizes a broad class of algorithms, with the hope that the resulting model will provide good accuracy-complexity tradeoffs in the average sense. We illustrate this setup on the Quadratic Assignment Problem, a fundamental problem in Network Science. We observe that data-driven models based on Graph Neural Networks offer intriguingly good performance, even in regimes where standard relaxation based techniques appear to suffer.

stat.ML↗

arXiv2018

Optimality and Sub-optimality of PCA I: Spiked Random Matrix Models

A central problem of random matrix theory is to understand the eigenvalues of spiked random matrix models, introduced by Johnstone, in which a prominent eigenvector (or "spike") is planted into a random matrix. These distributions form natural statistical models for principal component analysis (PCA) problems throughout the sciences. Baik, Ben Arous and Peche showed that the spiked Wishart ensemble exhibits a sharp phase transition asymptotically: when the spike strength is above a critical threshold, it is possible to detect the presence of a spike based on the top eigenvalue, and below the threshold the top eigenvalue provides no information. Such results form the basis of our understanding of when PCA can detect a low-rank signal in the presence of noise. However, under structural assumptions on the spike, not all information is necessarily contained in the spectrum. We study the statistical limits of tests for the presence of a spike, including non-spectral tests. Our results leverage Le Cam's notion of contiguity, and include: i) For the Gaussian Wigner ensemble, we show that PCA achieves the optimal detection threshold for certain natural priors for the spike. ii) For any non-Gaussian Wigner ensemble, PCA is sub-optimal for detection. However, an efficient variant of PCA achieves the optimal threshold (for natural priors) by pre-transforming the matrix entries. iii) For the Gaussian Wishart ensemble, the PCA threshold is optimal for positive spikes (for natural priors) but this is not always the case for negative spikes.

math.ST↗

arXiv2018

Stochastic Block Model for Hypergraphs: Statistical limits and a semidefinite programming approach

We study the problem of community detection in a random hypergraph model which we call the stochastic block model for $k$-uniform hypergraphs ($k$-SBM). We investigate the exact recovery problem in $k$-SBM and show that a sharp phase transition occurs around a threshold: below the threshold it is impossible to recover the communities with non-vanishing probability, yet above the threshold there is an estimator which recovers the communities almost asymptotically surely. We also consider a simple, efficient algorithm for the exact recovery problem which is based on a semidefinite relaxation technique.

math.PR↗

arXiv2018

Community Detection in Hypergraphs, Spiked Tensor Models, and Sum-of-Squares

We study the problem of community detection in hypergraphs under a stochastic block model. Similarly to how the stochastic block model in graphs suggests studying spiked random matrices, our model motivates investigating statistical and computational limits of exact recovery in a certain spiked tensor model. In contrast with the matrix case, the spiked model naturally arising from community detection in hypergraphs is different from the one arising in the so-called tensor Principal Component Analysis model. We investigate the effectiveness of algorithms in the Sum-of-Squares hierarchy on these models. Interestingly, our results suggest that these two apparently similar models exhibit significantly different computational to statistical gaps.

cs.DS↗

arXiv2018

Optimal rates of estimation for multi-reference alignment

In this paper, we establish optimal rates of adaptive estimation of a vector in the multi-reference alignment model, a problem with important applications in fields such as signal processing, image processing, and computer vision, among others. We describe how this model can be viewed as a multivariate Gaussian mixture model under the constraint that the centers belong to the orbit of a group. This enables us to derive matching upper and lower bounds that feature an interesting dependence on the signal-to-noise ratio of the model. Both upper and lower bounds are articulated around a tight local control of Kullback-Leibler divergences that showcases the central role of moment tensors in this problem.

math.ST↗

arXiv2018

Notes on computational-to-statistical gaps: predictions using statistical physics

In these notes we describe heuristics to predict computational-to-statistical gaps in certain statistical problems. These are regimes in which the underlying statistical problem is information-theoretically possible although no efficient algorithm exists, rendering the problem essentially unsolvable for large instances. The methods we describe here are based on mature, albeit non-rigorous, tools from statistical physics. These notes are based on a lecture series given by the authors at the Courant Institute of Mathematical Sciences in New York City, on May 16th, 2017.

stat.ML↗