arXiv subjects

Anil Damle

Publications and source records attributed to Anil Damle.

At least 19 recordsLinked to original sources

arXiv2026

Estimating Hierarchically Rank Structured Covariance Matrices

We consider the problem of estimating a high-dimensional covariance matrix from a very limited number of samples. This problem is ubiquitous in computational fluid dynamics, where a small number of fluid snapshots must be used to construct a Gramian matrix determining a reduced-order model, as well as in computational geoscience, where a small ensemble of Earth system forecasts must be used to estimate the covariance matrix associated with the forecast uncertainty. It is common practice to regularize the small-sample covariance by imposing a "localization" structure that enforces a physically realistic correlation length scale, imposing a sparsity constraint, "shrinking" towards a prescribed target, or attenuating small correlations. We propose an alternate technique that regularizes the small-sample covariance by imposing hierarchical rank structure. Compared to regularization methods that assume sparsity such as spatial localization, hierarchical rank structure accommodates a wider range of covariance matrices, roughly corresponding to situations where long-range correlations vary more smoothly than short-range ones. It also results in a data-sparse matrix format that permits highly efficient matrix-vector products. We present theory and algorithms which show how to efficiently estimate a high-dimensional, hierarchically rank structured covariance matrix from limited samples. Through an error analysis and numerical experiments with a variety of model problems, we demonstrate that these techniques are effective at reducing sampling errors, and that in many cases they achieve smaller estimation error than conventional techniques.

stat.CO↗

arXiv2026

Computing Strong Rank-Revealing Factorizations for Matrices with Orthonormal Rows

We show that a pivoting strategy due to Stewart (based on work by Bischof) computes a strong rank-revealing factorization when applied to a matrix with orthonormal rows. When paired with the classical column selection algorithm of Golub, Klema, and Stewart (GKS) it helps achieve rank-$k$ approximation accuracy bounds and basis conditioning as good as those from applying a strong rank-revealing factorization directly to A. We then extend this framework in two directions: (1) providing analysis of GKS when only approximations of right singular vectors are available and (2) providing a randomized variant of the pivoting strategy for matrices with orthonormal rows that achieves the same theoretical guarantees but can return the desired subset two orders of magnitude faster than the deterministic variant.

math.NA↗

arXiv2026

Linear Systems and Eigenvalue Problems: Open Questions from a Simons Workshop

This document presents a series of open questions arising in matrix computations, i.e., the numerical solution of linear algebra problems. It is a result of working groups at the workshop Linear Systems and Eigenvalue Problems, which was organized at the Simons Institute for the Theory of Computing program on Complexity and Linear Algebra in Fall 2025. The complexity and numerical solution of linear algebra problems is a crosscutting area between theoretical computer science and numerical analysis. The value of the particular problem formulations here is that they were produced via discussions between researchers from both groups. The open questions are organized in five categories: iterative solvers for linear systems, eigenvalue computation, low-rank approximation, randomized sketching, and other areas including tensors, quantum systems, and matrix functions. (Updated to reflect the status of the open problems as of August 20, 2026.)

math.NA↗

arXiv2025

Collect, Commit, Expand: Efficient CPQR-Based Column Selection for Extremely Wide Matrices

Column-pivoted QR (CPQR) factorization is a computational primitive used in numerous applications that require selecting a small set of ``representative'' columns from a much larger matrix. These include applications in spectral clustering, model-order reduction, low-rank approximation, and computational quantum chemistry, where the matrix being factorized has a moderate number of rows but an extremely large number of columns. We describe a modification of the Golub-Businger algorithm which, for many matrices of this type, can perform CPQR-based column selection much more efficiently. This algorithm, which we call CCEQR, is based on a three-step ``collect, commit, expand'' strategy that limits the number of columns being manipulated, while also transferring more computational effort from level-2 BLAS to level-3. Unlike most CPQR algorithms that exploit level-3 BLAS, CCEQR is deterministic, and provably recovers a column permutation equivalent to the one computed by the Golub-Businger algorithm. Tests on spectral clustering and Wannier basis localization problems demonstrate that on appropriately structured problems, CCEQR can significantly outperform GEQP3.

math.NA↗

arXiv2024

Comparing large language models for supervised analysis of students' lab notes

Recent advancements in large language models (LLMs) hold significant promise in improving physics education research that uses machine learning. In this study, we compare the application of various models to perform large-scale analysis of written text grounded in a physics education research classification problem: identifying skills in students' typed lab notes through sentence-level labeling. Specifically, we use training data to fine-tune two different LLMs, BERT and LLaMA, and compare the performance of these models to both a traditional bag of words approach and a few-shot LLM (without fine-tuning).} We evaluate the models based on their resource use, performance metrics, and research outcomes when identifying skills in lab notes. We find that higher-resource models often, but not necessarily, perform better than lower-resource models. We also find that all models estimate similar trends in research outcomes, although the absolute values of the estimated measurements are not always within uncertainties of each other. We use the results to discuss relevant considerations for education researchers seeking to select a model type to use as a classifier.

physics.ed-ph↗

arXiv2024

Automating Variational Differentiation

Many problems in Physics and Chemistry are formulated as the minimization of a functional. Therefore, methods for solving these problems typically require differentiating maps whose input and/or output are functions -- commonly referred to as variational differentiation. Such maps are not addressed at the mathematical level by the chain rule, which underlies modern symbolic and algorithmic differentiation (AD) systems. Although there are algorithmic solutions such as tracing and reverse accumulation, they do not provide human readability and introduce strict programming constraints that bottleneck performance, especially in high-performance computing (HPC) environments. In this manuscript, we propose a new computer theoretic model of differentiation by combining the pullback of the $\mathbf{B}$ and $\mathbf{C}$ combinators from the combinatory logic. Unlike frameworks based on the chain rule, this model differentiates a minimal complete basis for the space of computable functions. Consequently, the model is capable of analytic backpropagation and variational differentiation while supporting complex numbers. To demonstrate the generality of this approach we build a system named CombDiff, which can differentiate nontrivial variational problems such as Hartree-Fock (HF) theory and multilayer perceptrons.

cs.MS↗

arXiv2024

How to reveal the rank of a matrix?

We study algorithms called rank-revealers that reveal a matrix's rank structure. Such algorithms form a fundamental component in matrix compression, singular value estimation, and column subset selection problems. While column-pivoted QR has been widely adopted due to its practicality, it is not always a rank-revealer. Conversely, Gaussian elimination (GE) with a pivoting strategy known as global maximum volume pivoting is guaranteed to estimate a matrix's singular values but its exponential complexity limits its interest to theory. We show that the concept of local maximum volume pivoting is a crucial and practical pivoting strategy for rank-revealers based on GE and QR. In particular, we prove that it is both necessary and sufficient; highlighting that all local solutions are nearly as good as the global one. This insight elevates Gu and Eisenstat's rank-revealing QR as an archetypal rank-revealer, and we implement a version that is observed to be at most $2\times$ more computationally expensive than CPQR. We unify the landscape of rank-revealers by considering GE and QR together and prove that the success of any pivoting strategy can be assessed by benchmarking it against a local maximum volume pivot.

math.NA↗

arXiv2024

STAT: Shrinking Transformers After Training

We present STAT: a simple algorithm to prune transformer models without any fine-tuning. STAT eliminates both attention heads and neurons from the network, while preserving accuracy by calculating a correction to the weights of the next layer. Each layer block in the network is compressed using a series of principled matrix factorizations that preserve the network structure. Our entire algorithm takes minutes to compress BERT, and less than three hours to compress models with 7B parameters using a single GPU. Using only several hundred data examples, STAT preserves the output of the network and improves upon existing gradient-free pruning methods. It is even competitive with methods that include significant fine-tuning. We demonstrate our method on both encoder and decoder architectures, including BERT, DistilBERT, and Llama-2 using benchmarks such as GLUE, Squad, WikiText2.

cs.LG↗

arXiv2023

Structure-Aware Analyses and Algorithms for Interpolative Decompositions

Low-rank approximation is a task of critical importance in modern science, engineering, and statistics. Many low-rank approximation algorithms, such as the randomized singular value decomposition (RSVD), project their input matrix into a subspace approximating the span of its leading singular vectors. Other algorithms compress their input into a small subset of representative rows or columns, leading to a so-called interpolative decomposition. This paper investigates how the accuracy of interpolative decompositions is affected by the structural properties of the input matrix being operated on, including how these properties affect the performance comparison between interpolative decompositions and RSVD. We also introduce a novel method of interpolative decomposition in the form of the randomized Golub-Klema-Stewart (RGKS) algorithm, which combines RSVD with a pivoting strategy for column subset selection. Through numerical experiments, we find that matrix structures including singular subspace geometry and singular spectrum decay play a significant role in determining the performance comparison between these different algorithms. We also prove inequalities which bound the error of a general interpolative decomposition in terms of these matrix structures. Lastly, we develop forms of these bounds specialized to RGKS while considering how randomization affects the approximation error of this algorithm.

math.NA↗

arXiv2023

An unambiguous and robust formulation for Wannier localization

We provide a new variational definition for the spread of an orbital under periodic boundary conditions (PBCs) that is continuous with respect to the gauge, consistent in the thermodynamic limit, well-suited to diffuse orbitals, and systematically adaptable to schemes computing localized Wannier functions. Existing definitions do not satisfy all these desiderata, partly because they depend on an "orbital center"-an ill-defined concept under PBCs. Based on this theoretical development, we showcase a robust and efficient (10x-70x fewer iterations) localization scheme across a range of materials.

cond-mat.mtrl-sci↗

arXiv2022

Model Preserving Compression for Neural Networks

After training complex deep learning models, a common task is to compress the model to reduce compute and storage demands. When compressing, it is desirable to preserve the original model's per-example decisions (e.g., to go beyond top-1 accuracy or preserve robustness), maintain the network's structure, automatically determine per-layer compression levels, and eliminate the need for fine tuning. No existing compression methods simultaneously satisfy these criteria $\unicode{x2014}$ we introduce a principled approach that does by leveraging interpolative decompositions. Our approach simultaneously selects and eliminates channels (analogously, neurons), then constructs an interpolation matrix that propagates a correction into the next layer, preserving the network's structure. Consequently, our method achieves good performance even without fine tuning and admits theoretical analysis. Our theoretical generalization bound for a one layer network lends itself naturally to a heuristic that allows our method to automatically choose per-layer sizes for deep networks. We demonstrate the efficacy of our approach with strong empirical performance on a variety of tasks, models, and datasets $\unicode{x2014}$ from simple one-hidden-layer networks to deep networks on ImageNet.

cs.LG↗

arXiv2022

Communication-efficient distributed eigenspace estimation with arbitrary node failures

We develop an eigenspace estimation algorithm for distributed environments with arbitrary node failures, where a subset of computing nodes can return structurally valid but otherwise arbitrarily chosen responses. Notably, this setting encompasses several important scenarios that arise in distributed computing and data-collection environments such as silent/soft errors, outliers or corrupted data at certain nodes, and adversarial responses. Our estimator builds upon and matches the performance of a recently proposed non-robust estimator up to an additive $\tilde{O}(σ\sqrtα)$ error, where $σ^2$ is the variance of the existing estimator and $α$ is the fraction of corrupted nodes.

stat.ML↗

arXiv2022

Linear Time Kernel Matrix Approximation via Hyperspherical Harmonics

We propose a new technique for constructing low-rank approximations of matrices that arise in kernel methods for machine learning. Our approach pairs a novel automatically constructed analytic expansion of the underlying kernel function with a data-dependent compression step to further optimize the approximation. This procedure works in linear time and is applicable to any isotropic kernel. Moreover, our method accepts the desired error tolerance as input, in contrast to prevalent methods which accept the rank as input. Experimental results show our approach compares favorably to the commonly used Nystrom method with respect to both accuracy for a given rank and computational time for a given accuracy across a variety of kernels, dimensions, and datasets. Notably, in many of these problem settings our approach produces near-optimal low-rank approximations. We provide an efficient open-source implementation of our new technique to complement our theoretical developments and experimental results.

cs.LG↗

arXiv2021

Data-driven Algorithms for signal processing with trigonometric rational functions

Rational approximation schemes for reconstructing periodic signals from samples with poorly separated spectral content are described. These methods are automatic and adaptive, requiring no tuning or manual parameter selection. Collectively, they form a framework for fitting trigonometric rational models to data that is robust to various forms of corruption, including additive Gaussian noise, perturbed sampling grids, and missing data. Our approach combines a variant of Prony's method with a modified version of the AAA algorithm. Using representations in both frequency and time space, a collection of algorithms is described for adaptively computing with trigonometric rationals. This includes procedures for differentiation, filtering, convolution, and more. A new MATLAB software system based on these algorithms is introduced. Its effectiveness is illustrated with synthetic and practical examples drawn from applications including biomedical monitoring, acoustic denoising, and feature detection.

math.NA↗

arXiv2021

Selected Columns of the Density Matrix in an Atomic Orbital Basis I: An Intrinsic and Non-Iterative Orbital Localization Scheme for the Occupied Space

We extend the selected columns of the density matrix (SCDM) methodology [J. Chem. Theory Comput. 2015, 11, 1463--1469]---a non-iterative procedure for generating localized occupied orbitals for condensed-phase systems---to the construction of local molecular orbitals (LMOs) in systems described using non-orthogonal atomic orbital (AO) basis sets. In particular, we introduce three different variants of SCDM (referred to as SCDM-M, SCDM-L, and SCDM-G) that can be used in conjunction with the standard AO basis sets. The SCDM-M and SCDM-L variants are based on the Mulliken and L{ö}wdin representations of the density matrix, and are tantamount to selecting a well-conditioned set of projected atomic orbitals (PAOs) and projected (symmetrically-) orthogonalized atomic orbitals (POAOs), respectively, as proto-LMOs. The SCDM-G variant leverages a real-space (grid) representation of the wavefunction to select a set of well-conditioned proto-LMOs. A detailed comparative analysis reveals that the LMOs generated by these three SCDM variants are robust, comparable in orbital locality to those produced with the iterative Boys or Pipek-Mezey (PM) localization schemes, and are agnostic towards any single orbital locality metric. Although all three SCDM variants are based on the density matrix, we find that the character of the generated LMOs can differ significantly between SCDM-M, SCDM-L, and SCDM-G. In this regard, only the grid-based SCDM-G procedure (like PM) generates LMOs that qualitatively preserve $σ\text{-}π$ symmetry and are well-aligned with chemical intuition. While the direct and standalone use of SCDM-generated LMOs should suffice for most applications, our findings also suggest that the use of these orbitals as an unbiased and cost-effective (initial) guess also has the potential to improve the convergence of iterative orbital localization schemes.

physics.chem-ph↗

arXiv2021

The Fast Kernel Transform

Kernel methods are a highly effective and widely used collection of modern machine learning algorithms. A fundamental limitation of virtually all such methods are computations involving the kernel matrix that naively scale quadratically (e.g., constructing the kernel matrix and matrix-vector multiplication) or cubically (solving linear systems) with the size of the data set $N.$ We propose the Fast Kernel Transform (FKT), a general algorithm to compute matrix-vector multiplications (MVMs) for datasets in moderate dimensions with quasilinear complexity. Typically, analytically grounded fast multiplication methods require specialized development for specific kernels. In contrast, our scheme is based on auto-differentiation and automated symbolic computations that leverage the analytical structure of the underlying kernel. This allows the FKT to be easily applied to a broad class of kernels, including Gaussian, Matern, and Rational Quadratic covariance functions and physically motivated Green's functions, including those of the Laplace and Helmholtz equations. Furthermore, the FKT maintains a high, quantifiable, and controllable level of accuracy -- properties that many acceleration methods lack. We illustrate the efficacy and versatility of the FKT by providing timing and accuracy benchmarks and by applying it to scale the stochastic neighborhood embedding (t-SNE) and Gaussian processes to large real-world data sets.

cs.LG↗

arXiv2021

Communication-efficient distributed eigenspace estimation

Distributed computing is a standard way to scale up machine learning and data science algorithms to process large amounts of data. In such settings, avoiding communication amongst machines is paramount for achieving high performance. Rather than distribute the computation of existing algorithms, a common practice for avoiding communication is to compute local solutions or parameter estimates on each machine and then combine the results; in many convex optimization problems, even simple averaging of local solutions can work well. However, these schemes do not work when the local solutions are not unique. Spectral methods are a collection of such problems, where solutions are orthonormal bases of the leading invariant subspace of an associated data matrix, which are only unique up to rotation and reflections. Here, we develop a communication-efficient distributed algorithm for computing the leading invariant subspace of a data matrix. Our algorithm uses a novel alignment scheme that minimizes the Procrustean distance between local solutions and a reference solution, and only requires a single round of communication. For the important case of principal component analysis (PCA), we show that our algorithm achieves a similar error rate to that of a centralized estimator. We present numerical experiments demonstrating the efficacy of our proposed algorithm for distributed PCA, as well as other problems where solutions exhibit rotational symmetry, such as node embeddings for graph data and spectral initialization for quadratic sensing.

stat.ML↗

arXiv2021

Parallel Skeletonization for Integral Equations in Evolving Multiply-Connected Domains

This paper presents a general method for applying hierarchical matrix skeletonization factorizations to the numerical solution of boundary integral equations with possibly rank-deficient integral operators. Rank-deficient operators arise in boundary integral approaches to elliptic partial differential equations with multiple boundary components, such as in the case of multiple vesicles in a viscous fluid flow. Our generalized skeletonization factorization retains the locality property afforded by the "proxy point method," and allows for a parallelized implementation where different processors work on different parts of the boundary simultaneously. Further, when the boundary undergoes local geometric perturbations (such as movement of an interior hole), the factorization can be recomputed efficiently with respect to the number of modified discretization nodes. We present an application that leverages a parallel implementation of skeletonization with updates in a shape optimization regime.

math.NA↗