Search

In this work, we develop and analyze a Gradient Descent (GD) based solution, called Alternating GD and Minimization (AltGDmin), for efficiently solving the low rank matrix completion (LRMC) in a federated setting. LRMC involves recovering an $n \times q$ rank-$r$ matrix $\Xstar$ from a subset of its entries when $r \ll \min(n,q)$. Our theoretical guarantees (iteration and sample complexity bounds) imply that AltGDmin is the most communication-efficient solution in a federated setting, is one of the fastest, and has the second best sample complexity among all iterative solutions to LRMC. In addition, we also prove two important corollaries. (a) We provide a guarantee for AltGDmin for solving the noisy LRMC problem. (b) We show how our lemmas can be used to provide an improved sample complexity guarantee for AltMin, which is the fastest centralized solution.

This paper focuses studies the following low rank + sparse (LR+S) column-wise compressive sensing problem. We aim to recover an $n \times q$ matrix, $\X^* =[ \x_1^*, \x_2^*, \cdots , \x_q^*]$ from $m$ independent linear projections of each of its $q$ columns, given by $\y_k :=\A_k\x_k^*$, $k \in [q]$. Here, $\y_k$ is an $m$-length vector with $m < n$. We assume that the matrix $\X^*$ can be decomposed as $\X^*=\L^*+\S^*$, where $\L^*$ is a low rank matrix of rank $r << \min(n,q)$ and $\S^*$ is a sparse matrix. Each column of $\S$ contains $\rho$ non-zero entries. The matrices $\A_k$ are known and mutually independent for different $k$. To address this recovery problem, we propose a novel fast GD-based solution called AltGDmin-LR+S, which is memory and communication efficient. We numerically evaluate its performance by conducting a detailed simulation-based study., Comment: 6 pages, 2 figures, conference

This work considers two related learning problems in a federated attack prone setting: federated principal components analysis (PCA) and federated low rank column-wise sensing (LRCS). The node attacks are assumed to be Byzantine which means that the attackers are omniscient and can collude. We introduce a novel provably Byzantine-resilient communication-efficient and sampleefficient algorithm, called Subspace-Median, that solves the PCA problem and is a key part of the solution for the LRCS problem. We also study the most natural Byzantine-resilient solution for federated PCA, a geometric median based modification of the federated power method, and explain why it is not useful. Our second main contribution is a complete alternating gradient descent (GD) and minimization (altGDmin) algorithm for Byzantine-resilient horizontally federated LRCS and sample and communication complexity guarantees for it. Extensive simulation experiments are used to corroborate our theoretical guarantees. The ideas that we develop for LRCS are easily extendable to other LR recovery problems as well., Comment: 36 pages

This note provides a significantly simpler and shorter proof of our sample complexity guarantee for solving the low rank column-wise sensing problem using the Alternating Gradient Descent (GD) and Minimization (AltGDmin) algorithm. AltGDmin was developed and analyzed for solving this problem in our recent work. We also provide an improved guarantee.

This work develops a novel set of algorithms, alternating Gradient Descent (GD) and minimization for MRI (altGDmin-MRI1 and altGDmin-MRI2), for accelerated dynamic MRI by assuming an approximate low-rank (LR) model on the matrix formed by the vectorized images of the sequence. The LR model itself is well-known in the MRI literature; our contribution is the novel GD-based algorithms which are much faster, memory efficient, and general compared with existing work; and careful use of a 3-level hierarchical LR model. By general, we mean that, with a single choice of parameters, our method provides accurate reconstructions for multiple accelerated dynamic MRI applications, multiple sampling rates and sampling schemes. We show that our methods outperform many of the popular existing approaches while also being faster than all of them, on average. This claim is based on comparisons on 8 different retrospectively under sampled multi-coil dynamic MRI applications, sampled using either 1D Cartesian or 2D pseudo radial under sampling, at multiple sampling rates. Evaluations on some prospectively under sampled datasets are also provided. Our second contribution is a mini-batch subspace tracking extension that can process new measurements and return reconstructions within a short delay after they arrive. The recovery algorithm itself is also faster than its batch counterpart., Comment: 16 Pages (Including Appendix), 9 Figures, 3 Tables. arXiv admin note: substantial text overlap with arXiv:2206.13618

A plethora of modern machine learning tasks require the utilization of large-scale distributed clusters as a critical component of the training pipeline. However, abnormal Byzantine behavior of the worker nodes can derail the training and compromise the quality of the inference. Such behavior can be attributed to unintentional system malfunctions or orchestrated attacks; as a result, some nodes may return arbitrary results to the parameter server (PS) that coordinates the training. Recent work considers a wide range of attack models and has explored robust aggregation and/or computational redundancy to correct the distorted gradients. In this work, we consider attack models ranging from strong ones: $q$ omniscient adversaries with full knowledge of the defense protocol that can change from iteration to iteration to weak ones: $q$ randomly chosen adversaries with limited collusion abilities which only change every few iterations at a time. Our algorithms rely on redundant task assignments coupled with detection of adversarial behavior. We also show the convergence of our method to the optimal point under common assumptions and settings considered in literature. For strong attacks, we demonstrate a reduction in the fraction of distorted gradients ranging from 16%-99% as compared to the prior state-of-the-art. Our top-1 classification accuracy results on the CIFAR-10 data set demonstrate 25% advantage in accuracy (averaged over strong and weak scenarios) under the most sophisticated attacks compared to state-of-the-art methods., Comment: 21 pages, 17 figures, 6 tables. The material in this work appeared in part at arXiv:2108.02416 which has been published at the 2022 IEEE International Symposium on Information Theory

This work develops a provably accurate fully-decentralized alternating projected gradient descent (GD) algorithm for recovering a low rank (LR) matrix from mutually independent projections of each of its columns, in a fast and communication-efficient fashion. To our best knowledge, this work is the first attempt to develop a provably correct decentralized algorithm (i) for any problem involving the use of an alternating projected GD algorithm; (ii) and for any problem in which the constraint set to be projected to is a non-convex set.

In this work we develop a fully decentralized, federated, and fast solution to the recently studied Low Rank Compressive Sensing (LRCS) problem: recover an nxq low-rank matrix from column-wise linear projections. An important application where this problem occurs, and a decentralized solution is desirable is in federated sketching: efficiently compressing the vast amounts of distributed images/videos generated by smartphones and various other devices while respecting the users' privacy. Images from different devices, once grouped by category, are similar and hence the matrix formed by the vectorized images of a certain category is well-modeled as being low rank. Suppose there are p nodes (say p smartphones), and each store a subset of the sketches of its images. We develop a decentralized projected gradient descent (GD) based approach to jointly reconstruct the images of all the phones/users from their respective stored sketches. The algorithm is such that the phones/users never share their raw data but only summaries of this data with the other phones at each algorithm iteration. Also, the reconstructed images of user g are obtained only locally. Other users cannot reconstruct them. Only the column span of the matrix is reconstructed globally. By "decentralized" we mean that there is no central node to which all nodes are connected and thus the only way to aggregate the summaries from the various nodes is by use of an iterative consensus algorithm that eventually provides an estimate of the aggregate at each node, as long as the network is strongly connected. We validated the effectiveness of our algorithm via extensive simulation experiments.

We study the following lesser-known low rank (LR) recovery problem: recover an $n \times q$ rank-$r$ matrix, $X^* =[x^*_1 , x^*_2, ..., x^*_q]$, with $r \ll \min(n,q)$, from $m$ independent linear projections of each of its $q$ columns, i.e., from $y_k := A_k x^*_k , k \in [q]$, when $y_k$ is an $m$-length vector with $m < n$. The matrices $A_k$ are known and mutually independent for different $k$. We introduce a novel gradient descent (GD) based solution called AltGD-Min. We show that, if the $A_k$s are i.i.d. with i.i.d. Gaussian entries, and if the right singular vectors of $X^*$ satisfy the incoherence assumption, then $\epsilon$-accurate recovery of $X^*$ is possible with order $(n+q) r^2 \log(1/\epsilon)$ total samples and order $ mq nr \log (1/\epsilon)$ time. Compared with existing work, this is the fastest solution. For $\epsilon < r^{1/4}$, it also has the best sample complexity. A simple extension of AltGD-Min also provably solves LR Phase Retrieval, which is a magnitude-only generalization of the above problem. AltGD-Min factorizes the unknown $X$ as $X = UB$ where $U$ and $B$ are matrices with $r$ columns and rows respectively. It alternates between a (projected) GD step for updating $U$, and a minimization step for updating $B$. Its each iteration is as fast as that of regular projected GD because the minimization over $B$ decouples column-wise. At the same time, we can prove exponential error decay for it, which we are unable to for projected GD. Finally, it can also be efficiently federated with a communication cost of only $nr$ per node, instead of $nq$ for projected GD., Comment: To appear in IEEE Transactions on Information Theory (T-IT)

Phase retrieval (PR), also sometimes referred to as quadratic sensing, is a problem that occurs in numerous signal and image acquisition domains ranging from optics, X-ray crystallography, Fourier ptychography, sub-diffraction imaging, and astronomy. In each of these domains, the physics of the acquisition system dictates that only the magnitude (intensity) of certain linear projections of the signal or image can be measured. Without any assumptions on the unknown signal, accurate recovery necessarily requires an over-complete set of measurements. The only way to reduce the measurements/sample complexity is to place extra assumptions on the unknown signal/image. A simple and practically valid set of assumptions is obtained by exploiting the structure inherently present in many natural signals or sequences of signals. Two commonly used structural assumptions are (i) sparsity of a given signal/image or (ii) a low rank model on the matrix formed by a set, e.g., a time sequence, of signals/images. Both have been explored for solving the PR problem in a sample-efficient fashion. This article describes this work, with a focus on non-convex approaches that come with sample complexity guarantees under simple assumptions. We also briefly describe other different types of structural assumptions that have been used in recent literature., Comment: to appear in IEEE Signal Processing Magazine (Special Issue on Non-Convex Optimization for Signal Processing and Machine Learning)

This work studies the robust subspace tracking (ST) problem. Robust ST can be simply understood as a (slow) time-varying subspace extension of robust PCA. It assumes that the true data lies in a low-dimensional subspace that is either fixed or changes slowly with time. The goal is to track the changing subspaces over time in the presence of additive sparse outliers and to do this quickly (with a short delay). We introduce a "fast" mini-batch robust ST solution that is provably correct under mild assumptions. Here "fast" means two things: (i) the subspace changes can be detected and the subspaces can be tracked with near-optimal delay, and (ii) the time complexity of doing this is the same as that of simple (non-robust) PCA. Our main result assumes piecewise constant subspaces (needed for identifiability), but we also provide a corollary for the case when there is a little change at each time. A second contribution is a novel non-asymptotic guarantee for PCA in linearly data-dependent noise. An important setting where this is useful is for linearly data dependent noise that is sparse with support that changes enough over time. The analysis of the subspace update step of our proposed robust ST solution uses this result., Comment: To appear in IEEE Journal of Special Areas in Information Theory

This work studies the Low Rank Phase Retrieval (LRPR) problem: recover an $n \times q$ rank-$r$ matrix $X^*$ from $y_k = |A_k^\top x^*_k|$, $k=1, 2,..., q$, when each $y_k$ is an m-length vector containing independent phaseless linear projections of $x^*_k$. The different matrices $A_k$ are i.i.d. and each contains i.i.d. standard Gaussian entries. We obtain an improved guarantee for AltMinLowRaP, which is an Alternating Minimization solution to LRPR that was introduced and studied in our recent work. As long as the right singular vectors of $X^*$ satisfy the incoherence assumption, we can show that the AltMinLowRaP estimate converges geometrically to $X^*$ if the total number of measurements $mq \gtrsim nr^2 (r + \log(1/\epsilon))$. In addition, we also need $m \gtrsim max(r, \log q, \log n)$ because of the specific asymmetric nature of our problem. Compared to our recent work, we improve the sample complexity of the AltMin iterations by a factor of $r^2$, and that of the initialization by a factor of $r$. We also extend our result to the noisy case; we prove stability to corruption by small additive noise., Comment: Revised for IEEE Trans. Info. Th

In this work we study the problem of Subspace Tracking with missing data (ST-miss) and outliers (Robust ST-miss). We propose a novel algorithm, and provide a guarantee for both these problems. Unlike past work on this topic, the current work does not impose the piecewise constant subspace change assumption. Additionally, the proposed algorithm is much simpler (uses fewer parameters) than our previous work. Secondly, we extend our approach and its analysis to provably solving these problems when the data is federated and when the over-air data communication modality is used for information exchange between the $K$ peer nodes and the center. We validate our theoretical claims with extensive numerical experiments., Comment: To appear in IEEE Transactions on Signal Processing. changes to writing; more general result provided from which previous result follows as special case

Distributed matrix computations -- matrix-matrix or matrix-vector multiplications -- are well-recognized to suffer from the problem of stragglers (slow or failed worker nodes). Much of prior work in this area is (i) either sub-optimal in terms of its straggler resilience, or (ii) suffers from numerical problems, i.e., there is a blow-up of round-off errors in the decoded result owing to the high condition numbers of the corresponding decoding matrices. Our work presents convolutional coding approach to this problem that removes these limitations. It is optimal in terms of its straggler resilience, and has excellent numerical robustness as long as the workers' storage capacity is slightly higher than the fundamental lower bound. Moreover, it can be decoded using a fast peeling decoder that only involves add/subtract operations. Our second approach has marginally higher decoding complexity than the first one, but allows us to operate arbitrarily close to the lower bound. Its numerical robustness can be theoretically quantified by deriving a computable upper bound on the worst case condition number over all possible decoding matrices by drawing connections with the properties of large Toeplitz matrices. All above claims are backed up by extensive experiments done on the AWS cloud platform., Comment: 31 pages

We study the Low Rank Phase Retrieval (LRPR) problem defined as follows: recover an $n \times q$ matrix $X^*$ of rank $r$ from a different and independent set of $m$ phaseless (magnitude-only) linear projections of each of its columns. To be precise, we need to recover $X^*$ from $y_k := |A_k{}' x^*_k|, k=1,2,\dots, q$ when the measurement matrices $A_k$ are mutually independent. Here $y_k$ is an $m$ length vector, $A_k$ is an $n \times m$ matrix, and $'$ denotes matrix transpose. The question is when can we solve LRPR with $m \ll n$? A reliable solution can enable fast and low-cost phaseless dynamic imaging, e.g., Fourier ptychographic imaging of live biological specimens. In this work, we develop the first provably correct approach for solving this LRPR problem. Our proposed algorithm, Alternating Minimization for Low-Rank Phase Retrieval (AltMinLowRaP), is an AltMin based solution and hence is also provably fast (converges geometrically). Our guarantee shows that AltMinLowRaP solves LRPR to $\epsilon$ accuracy, with high probability, as long as $m q \ge C n r^4 \log(1/\epsilon)$, the matrices $A_k$ contain i.i.d. standard Gaussian entries, and the right singular vectors of $X^*$ satisfy the incoherence assumption from matrix completion literature. Here $C$ is a numerical constant that only depends on the condition number of $X^*$ and on its incoherence parameter. Its time complexity is only $ C mq nr \log^2(1/\epsilon)$. Since even the linear (with phase) version of the above problem is not fully solved, the above result is also the first complete solution and guarantee for the linear case. Finally, we also develop a simple extension of our results for the dynamic LRPR setting., Comment: A short version of this work is in ICML 2019, this longer version is published in IEEE Trans. Info. Th on March 2020. Fixing minor but important errors in Lemmas 3.10, 3.11, 3.12 statements and in proof of the Term1 bound. No change to Theorem statement

We study the problem of subspace tracking in the presence of missing data (ST-miss). In recent work, we studied a related problem called robust ST. In this work, we show that a simple modification of our robust ST solution also provably solves ST-miss and robust ST-miss. To our knowledge, our result is the first `complete' guarantee for ST-miss. This means that we can prove that under assumptions on only the algorithm inputs, the output subspace estimates are close to the true data subspaces at all times. Our guarantees hold under mild and easily interpretable assumptions, and allow the underlying subspace to change with time in a piecewise constant fashion. In contrast, all existing guarantees for ST are partial results and assume a fixed unknown subspace. Extensive numerical experiments are shown to back up our theoretical claims. Finally, our solution can be interpreted as a provably correct mini-batch and memory-efficient solution to low-rank Matrix Completion (MC)., Comment: Writing changes; includes a detailed discussion of noise analysis; contains discussion for Matrix Completion; Accepted to IEEE Transactions on Signal Processing

This work takes the first steps towards solving the "phaseless subspace tracking" (PST) problem. PST involves recovering a time sequence of signals (or images) from phaseless linear projections of each signal under the following structural assumption: the signal sequence is generated from a much lower dimensional subspace (than the signal dimension) and this subspace can change over time, albeit gradually. It can be simply understood as a dynamic (time-varying subspace) extension of the low-rank phase retrieval problem studied in recent work., Comment: To be appeared in GlobalSIP 2018

Principal Components Analysis (PCA) is one of the most widely used dimension reduction techniques. Robust PCA (RPCA) refers to the problem of PCA when the data may be corrupted by outliers. Recent work by Cand{\`e}s, Wright, Li, and Ma defined RPCA as a problem of decomposing a given data matrix into the sum of a low-rank matrix (true data) and a sparse matrix (outliers). The column space of the low-rank matrix then gives the PCA solution. This simple definition has lead to a large amount of interesting new work on provably correct, fast, and practical solutions to RPCA. More recently, the dynamic (time-varying) version of the RPCA problem has been studied and a series of provably correct, fast, and memory efficient tracking solutions have been proposed. Dynamic RPCA (or robust subspace tracking) is the problem of tracking data lying in a (slowly) changing subspace while being robust to sparse outliers. This article provides an exhaustive review of the last decade of literature on RPCA and its dynamic counterpart (robust subspace tracking), along with describing their theoretical guarantees, discussing the pros and cons of various approaches, and providing empirical comparisons of performance and speed. A brief overview of the (low-rank) matrix completion literature is also provided (the focus is on works not discussed in other recent reviews). This refers to the problem of completing a low-rank matrix when only a subset of its entries are observed. It can be interpreted as a simpler special case of RPCA in which the indices of the outlier corrupted entries are known., Comment: To appear in Proceedings of the IEEE, Special Issue on Rethinking PCA for Modern Datasets. arXiv admin note: text overlap with arXiv:1711.09492

In this work, we study the robust subspace tracking (RST) problem and obtain one of the first two provable guarantees for it. The goal of RST is to track sequentially arriving data vectors that lie in a slowly changing low-dimensional subspace, while being robust to corruption by additive sparse outliers. It can also be interpreted as a dynamic (time-varying) extension of robust PCA (RPCA), with the minor difference that RST also requires a short tracking delay. We develop a recursive projected compressive sensing algorithm that we call Nearly Optimal RST via ReProCS (ReProCS-NORST) because its tracking delay is nearly optimal. We prove that NORST solves both the RST and the dynamic RPCA problems under weakened standard RPCA assumptions, two simple extra assumptions (slow subspace change and most outlier magnitudes lower bounded), and a few minor assumptions. Our guarantee shows that NORST enjoys a near optimal tracking delay of $O(r \log n \log(1/\epsilon))$. Its required delay between subspace change times is the same, and its memory complexity is $n$ times this value. Thus both these are also nearly optimal. Here $n$ is the ambient space dimension, $r$ is the subspaces' dimension, and $\epsilon$ is the tracking accuracy. NORST also has the best outlier tolerance compared with all previous RPCA or RST methods, both theoretically and empirically (including for real videos), without requiring any model on how the outlier support is generated. This is possible because of the extra assumptions it uses., Comment: A [short version](http://proceedings.mlr.press/v80/narayanamurthy18a.html) will be presented at ICML 2018 (Long Talk). arXiv admin note: text overlap with arXiv:1803.00651

PCA is one of the most widely used dimension reduction techniques. A related easier problem is "subspace learning" or "subspace estimation". Given relatively clean data, both are easily solved via singular value decomposition (SVD). The problem of subspace learning or PCA in the presence of outliers is called robust subspace learning or robust PCA (RPCA). For long data sequences, if one tries to use a single lower dimensional subspace to represent the data, the required subspace dimension may end up being quite large. For such data, a better model is to assume that it lies in a low-dimensional subspace that can change over time, albeit gradually. The problem of tracking such data (and the subspaces) while being robust to outliers is called robust subspace tracking (RST). This article provides a magazine-style overview of the entire field of robust subspace learning and tracking. In particular solutions for three problems are discussed in detail: RPCA via sparse+low-rank matrix decomposition (S+LR), RST via S+LR, and "robust subspace recovery (RSR)". RSR assumes that an entire data vector is either an outlier or an inlier. The S+LR formulation instead assumes that outliers occur on only a few data vector indices and hence are well modeled as sparse corruptions., Comment: To appear, IEEE Signal Processing Magazine, July 2018

Video denoising refers to the problem of removing "noise" from a video sequence. Here the term "noise" is used in a broad sense to refer to any corruption or outlier or interference that is not the quantity of interest. In this work, we develop a novel approach to video denoising that is based on the idea that many noisy or corrupted videos can be split into three parts - the "low-rank layer", the "sparse layer", and a small residual (which is small and bounded). We show, using extensive experiments, that our denoising approach outperforms the state-of-the-art denoising algorithms., Comment: Shorter version with title "Video Denoising via Online Sparse and Low-rank Matrix Decomposition" appeared in Statistical Signal Processing Workshop (SSP) 2016

This work obtains novel finite sample guarantees for Principal Component Analysis (PCA). These hold even when the corrupting noise is non-isotropic, and a part (or all of it) is data-dependent. Because of the latter, in general, the noise and the true data are correlated. The results in this work are a significant improvement over those given in our earlier work where this "correlated-PCA" problem was first studied. In fact, in certain regimes, our results imply that the sample complexity required to achieve subspace recovery error that is a constant fraction of the noise level is near-optimal. Useful corollaries of our result include guarantees for PCA in sparse data-dependent noise and for PCA with missing data. An important application of the former is in proving correctness of the subspace update step of a popular online algorithm for dynamic robust PCA.

Dynamic robust PCA refers to the dynamic (time-varying) extension of robust PCA (RPCA). It assumes that the true (uncorrupted) data lies in a low-dimensional subspace that can change with time, albeit slowly. The goal is to track this changing subspace over time in the presence of sparse outliers. We develop and study a novel algorithm, that we call simple-ReProCS, based on the recently introduced Recursive Projected Compressive Sensing (ReProCS) framework. Our work provides the first guarantee for dynamic RPCA that holds under weakened versions of standard RPCA assumptions, slow subspace change and a lower bound assumption on most outlier magnitudes. Our result is significant because (i) it removes the strong assumptions needed by the two previous complete guarantees for ReProCS-based algorithms; (ii) it shows that it is possible to achieve significantly improved outlier tolerance, compared with all existing RPCA or dynamic RPCA solutions by exploiting the above two simple extra assumptions; and (iii) it proves that simple-ReProCS is online (after initialization), fast, and, has near-optimal memory complexity., Comment: Minor writing edits. The paper has been accepted to IEEE Transactions on Information Theory

We study Principal Component Analysis (PCA) in a setting where a part of the corrupting noise is data-dependent and, as a result, the noise and the true data are correlated. Under a bounded-ness assumption on the true data and the noise, and a simple assumption on data-noise correlation, we obtain a nearly optimal sample complexity bound for the most commonly used PCA solution, singular value decomposition (SVD). This bound is a significant improvement over the bound obtained by Vaswani and Guo in recent work (NIPS 2016) where this "correlated-PCA" problem was first studied; and it holds under a significantly weaker data-noise correlation assumption than the one used for this earlier result., Comment: made significant changes to paper. The completed new version is available at https://arxiv.org/abs/1709.06255

In this short note, we give a very simple but useful generalization of a result of Vershynin (Theorem 5.39 of [1]) for a random matrix with independent sub-Gaussian rows. We also explain with an example where our generalization is useful.

Given a matrix of observed data, Principal Components Analysis (PCA) computes a small number of orthogonal directions that contain most of its variability. Provably accurate solutions for PCA have been in use for a long time. However, to the best of our knowledge, all existing theoretical guarantees for it assume that the data and the corrupting noise are mutually independent, or at least uncorrelated. This is valid in practice often, but not always. In this paper, we study the PCA problem in the setting where the data and noise can be correlated. Such noise is often also referred to as "data-dependent noise". We obtain a correctness result for the standard eigenvalue decomposition (EVD) based solution to PCA under simple assumptions on the data-noise correlation. We also develop and analyze a generalization of EVD, cluster-EVD, that improves upon EVD in certain regimes., Comment: This paper has been withdrawn. It should have been a replacement of arXiv 1608.04320

We develop two iterative algorithms for solving the low rank phase retrieval (LRPR) problem. LRPR refers to recovering a low-rank matrix $\X$ from magnitude-only (phaseless) measurements of random linear projections of its columns. Both methods consist of a spectral initialization step followed by an iterative algorithm to maximize the observed data likelihood. We obtain sample complexity bounds for our proposed initialization approach to provide a good approximation of the true $\X$. When the rank is low enough, these bounds are significantly lower than what existing single vector phase retrieval algorithms need. Via extensive experiments, we show that the same is also true for the proposed complete algorithms., Comment: To appear in IEEE Trans. Signal Processing, 2017

In this article, we review the literature on design and analysis of recursive algorithms for reconstructing a time sequence of sparse signals from compressive measurements. The signals are assumed to be sparse in some transform domain or in some dictionary. Their sparsity patterns can change with time, although, in many practical applications, the changes are gradual. An important class of applications where this problem occurs is dynamic projection imaging, e.g., dynamic magnetic resonance imaging (MRI) for real-time medical applications such as interventional radiology, or dynamic computed tomography., Comment: To appear in IEEE Trans. Signal Processing

In this work, we study the online robust principal components' analysis (RPCA) problem. In recent work, RPCA has been defined as a problem of separating a low-rank matrix (true data), $L$, and a sparse matrix (outliers), $S$, from their sum, $M:=L + S$. A more general version of this problem is to recover $L$ and $S$ from $M:=L + S + W$ where $W$ is the matrix of unstructured small noise/corruptions. An important application where this problem occurs is in video analytics in trying to separate sparse foregrounds (e.g., moving objects) from slowly changing backgrounds. While there has been a large amount of recent work on solutions and guarantees for the batch RPCA problem, the online problem is largely open."Online" RPCA is the problem of doing the above on-the-fly with the extra assumptions that the initial subspace is accurately known and that the subspace from which $l_t$ is generated changes slowly over time. We develop and study a novel "online" RPCA algorithm based on the recently introduced Recursive Projected Compressive Sensing (ReProCS) framework. Our algorithm improves upon the original ReProCS algorithm and it also returns even more accurate offline estimates. The key contribution of this work is a correctness result (complete performance guarantee) for this algorithm under reasonably mild assumptions. By using extra assumptions -- accurate initial subspace knowledge, slow subspace change, and clustered eigenvalues -- we are able to remove one important limitation of batch RPCA results and two key limitations of a recent result for ReProCS for online RPCA. To our knowledge, this work is among the first few correctness results for online RPCA. Most earlier results were only partial results, i.e., they required an assumption on intermediate algorithm estimates., Comment: A shorter version of this work will be presented at AISTATS 2016

This work studies two interrelated problems - online robust PCA (RPCA) and online low-rank matrix completion (MC). In recent work by Cand\`{e}s et al., RPCA has been defined as a problem of separating a low-rank matrix (true data), $L:=[\ell_1, \ell_2, \dots \ell_{t}, \dots , \ell_{t_{\max}}]$ and a sparse matrix (outliers), $S:=[x_1, x_2, \dots x_{t}, \dots, x_{t_{\max}}]$ from their sum, $M:=L+S$. Our work uses this definition of RPCA. An important application where both these problems occur is in video analytics in trying to separate sparse foregrounds (e.g., moving objects) and slowly changing backgrounds. While there has been a large amount of recent work on both developing and analyzing batch RPCA and batch MC algorithms, the online problem is largely open. In this work, we develop a practical modification of our recently proposed algorithm to solve both the online RPCA and online MC problems. The main contribution of this work is that we obtain correctness results for the proposed algorithms under mild assumptions. The assumptions that we need are: (a) a good estimate of the initial subspace is available (easy to obtain using a short sequence of background-only frames in video surveillance); (b) the $\ell_t$'s obey a `slow subspace change' assumption; (c) the basis vectors for the subspace from which $\ell_t$ is generated are dense (non-sparse); (d) the support of $x_t$ changes by at least a certain amount at least every so often; and (e) algorithm parameters are appropriately set, Comment: Presented at ISIT (IEEE Intnl. Symp. on Information Theory), 2015. Submitted to IEEE Transactions on Information Theory. This version: changes are in blue; the main changes are just to explain the model assumptions better (added based on ISIT reviewers' comments)

This work studies the problem of sequentially recovering a sparse vector $x_t$ and a vector from a low-dimensional subspace $l_t$ from knowledge of their sum $m_t = x_t + l_t$. If the primary goal is to recover the low-dimensional subspace where the $l_t$'s lie, then the problem is one of online or recursive robust principal components analysis (PCA). To the best of our knowledge, this is the first correctness result for online robust PCA. We prove that if the $l_t$'s obey certain denseness and slow subspace change assumptions, and the support of $x_t$ changes by at least a certain amount at least every so often, and some other mild assumptions hold, then with high probability, the support of $x_t$ will be recovered exactly, and the error made in estimating $x_t$ and $l_t$ will be small. An example of where such a problem might arise is in separating a sparse foreground and slowly changing dense background in a surveillance video., Comment: A shorter version of this appears in the Proceedings of ICASSP 2015. Please read arXiv:1409.3959 (Online Matrix Completion and Online Robust PCA, Proc. of ISIT 2015 and submitted to IEEE Trans. Info. Th.) for a strictly improved correctness result (it provides a performance guarantee for a more practical ReProCS algorithm under almost the same conditions as this work)

Online or recursive robust PCA can be posed as a problem of recovering a sparse vector, $S_t$, and a dense vector, $L_t$, which lies in a slowly changing low-dimensional subspace, from $M_t:= S_t + L_t$ on-the-fly as new data comes in. For initialization, it is assumed that an accurate knowledge of the subspace in which $L_0$ lies is available. In recent works, Qiu et al proposed and analyzed a novel solution to this problem called recursive projected compressed sensing or ReProCS. In this work, we relax one limiting assumption of Qiu et al's result. Their work required that the $L_t$'s be mutually independent over time. However this is not a practical assumption, e.g., in the video application, $L_t$ is the background image sequence and one would expect it to be correlated over time. In this work we relax this and allow the $L_t$'s to follow an autoregressive model. We are able to show that under mild assumptions and under a denseness assumption on the unestimated part of the changed subspace, with high probability (w.h.p.), ReProCS can exactly recover the support set of $S_t$ at all times; the reconstruction errors of both $S_t$ and $L_t$ are upper bounded by a time invariant and small value; and the subspace recovery error decays to a small value within a finite delay of a subspace change time. Because the last assumption depends on an algorithm estimate, this result cannot be interpreted as a correctness result but only a useful step towards it., Comment: long version of conference paper

In recent work, robust Principal Components Analysis (PCA) has been posed as a problem of recovering a low-rank matrix $\mathbf{L}$ and a sparse matrix $\mathbf{S}$ from their sum, $\mathbf{M}:= \mathbf{L} + \mathbf{S}$ and a provably exact convex optimization solution called PCP has been proposed. This work studies the following problem. Suppose that we have partial knowledge about the column space of the low rank matrix $\mathbf{L}$. Can we use this information to improve the PCP solution, i.e. allow recovery under weaker assumptions? We propose here a simple but useful modification of the PCP idea, called modified-PCP, that allows us to use this knowledge. We derive its correctness result which shows that, when the available subspace knowledge is accurate, modified-PCP indeed requires significantly weaker incoherence assumptions than PCP. Extensive simulations are also used to illustrate this. Comparisons with PCP and other existing work are shown for a stylized real application as well. Finally, we explain how this problem naturally occurs in many applications involving time series data, i.e. in what is called the online or recursive robust PCA problem. A corollary for this case is also given., Comment: 19 pages, 9 figures, submitted to IEEE Transaction on Signal Processing

This work studies the recursive robust principal components analysis (PCA) problem. If the outlier is the signal-of-interest, this problem can be interpreted as one of recursively recovering a time sequence of sparse vectors, $S_t$, in the presence of large but structured noise, $L_t$. The structure that we assume on $L_t$ is that $L_t$ is dense and lies in a low dimensional subspace that is either fixed or changes "slowly enough". A key application where this problem occurs is in video surveillance where the goal is to separate a slowly changing background ($L_t$) from moving foreground objects ($S_t$) on-the-fly. To solve the above problem, in recent work, we introduced a novel solution called Recursive Projected CS (ReProCS). In this work we develop a simple modification of the original ReProCS idea and analyze it. This modification assumes knowledge of a subspace change model on the $L_t$'s. Under mild assumptions and a denseness assumption on the unestimated part of the subspace of $L_t$ at various times, we show that, with high probability (w.h.p.), the proposed approach can exactly recover the support set of $S_t$ at all times; and the reconstruction errors of both $S_t$ and $L_t$ are upper bounded by a time-invariant and small value. In simulation experiments, we observe that the last assumption holds as long as there is some support change of $S_t$ every few frames., Comment: This paper has been withdrawn because it is available at arXiv:1211.3754

This paper designs and evaluates a practical algorithm, called practical recursive projected compressive sensing (Prac-ReProCS), for recovering a time sequence of sparse vectors $S_t$ and a time sequence of dense vectors $L_t$ from their sum, $M_t:= S_t + L_t$, when any subsequence of the $L_t$'s lies in a slowly changing low-dimensional subspace. A key application where this problem occurs is in video layering where the goal is to separate a video sequence into a slowly changing background sequence and a sparse foreground sequence that consists of one or more moving regions/objects. Prac-ReProCS is a practical modification of its theoretical counterpart which was analyzed in our recent work. Experimental comparisons demonstrating the advantage of the approach for both simulated and real videos are shown. Extension to the undersampled case is also developed., Comment: Accepted by IEEE Transactions on Signal Processing. The title has been changed from "Practical ReProCS for Separating Sparse and Low-dimensional Signal Sequences from their Sum" to "An Online Algorithm for Separating Sparse and Low-dimensional Signal Sequences from their Sum"

In this work, we obtain performance guarantees for modified-CS and for its improved version, modified-CS-Add-LS-Del, for recursive reconstruction of a time sequence of sparse signals from a reduced set of noisy measurements available at each time. Under mild assumptions, we show that the support recovery error of both algorithms is bounded by a time-invariant and small value at all times. The same is also true for the reconstruction error. Under a slow support change assumption, (i) the support recovery error bound is small compared to the support size; and (ii) our results hold under weaker assumptions on the number of measurements than what $\ell_1$ minimization for noisy data needs. We first give a general result that only assumes a bound on support size, number of support changes and number of small magnitude nonzero entries at each time. Later, we specialize the main idea of these results for two sets of signal change assumptions that model the class of problems in which a new element that is added to the support either gets added at a large initial magnitude or its magnitude slowly increases to a large enough value within a finite delay. Simulation experiments are shown to back up our claims., Comment: Accepted to IEEE Transaction on Information Theory. arXiv admin note: substantial text overlap with arXiv:1104.2108

We study the problem of recursively recovering a time sequence of sparse vectors, St, from measurements Mt := St + Lt that are corrupted by structured noise Lt which is dense and can have large magnitude. The structure that we require is that Lt should lie in a low dimensional subspace that is either fixed or changes "slowly enough"; and the eigenvalues of its covariance matrix are "clustered". We do not assume any model on the sequence of sparse vectors. Their support sets and their nonzero element values may be either independent or correlated over time (usually in many applications they are correlated). The only thing required is that there be some support change every so often. We introduce a novel solution approach called Recursive Projected Compressive Sensing with cluster-PCA (ReProCS-cPCA) that addresses some of the limitations of earlier work. Under mild assumptions, we show that, with high probability, ReProCS-cPCA can exactly recover the support set of St at all times; and the reconstruction errors of both St and Lt are upper bounded by a time-invariant and small value.

This work studies the recursive robust principal components' analysis(PCA) problem. Here, "robust" refers to robustness to both independent and correlated sparse outliers. If the outlier is the signal-of-interest, this problem can be interpreted as one of recursively recovering a time sequence of sparse vectors, St, in the presence of large but structured noise, Lt. The structure that we assume on Lt is that Lt is dense and lies in a low dimensional subspace that is either fixed or changes "slowly enough". A key application where this problem occurs is in video surveillance where the goal is to separate a slowly changing background (Lt) from moving foreground objects (St) on-the-fly. To solve the above problem, we introduce a novel solution called Recursive Projected CS (ReProCS). Under mild assumptions, we show that, with high probability (w.h.p.), ReProCS can exactly recover the support set of St at all times; and the reconstruction errors of both St and Lt are upper bounded by a time-invariant and small value at all times., Comment: This version was accepted for publication in IEEE Transactions on Information Theory, August 2014

In this correspondence, we obtain exact recovery conditions for regularized modified basis pursuit (reg-mod-BP) and discuss when the obtained conditions are weaker than those for modified-CS or for basis pursuit (BP). The discussion is also supported by simulation comparisons. Reg-mod-BP provides a solution to the sparse recovery problem when both an erroneous estimate of the signal's support, denoted by $T$, and an erroneous estimate of the signal values on $T$ are available., Comment: 17 pages

This work studies the recursive robust principal components' analysis (PCA) problem. Here, "robust" refers to robustness to both independent and correlated sparse outliers, although we focus on the latter. A key application where this problem occurs is in video surveillance where the goal is to separate a slowly changing background from moving foreground objects on-the-fly. The background sequence is well modeled as lying in a low dimensional subspace, that can gradually change over time, while the moving foreground objects constitute the correlated sparse outliers. In this and many other applications, the foreground is an outlier for PCA but is actually the "signal of interest" for the application; where as the background is the corruption or noise. Thus our problem can also be interpreted as one of recursively recovering a time sequence of sparse signals in the presence of large but spatially correlated noise. This work has two key contributions. First, we provide a new way of looking at this problem and show how a key part of our solution strategy involves solving a noisy compressive sensing (CS) problem. Second, we show how we can utilize the correlation of the outliers to our advantage in order to even deal with very large support sized outliers. The main idea is as follows. The correlation model applied to the previous support estimate helps predict the current support. This prediction serves as "partial support knowledge" for solving the modified-CS problem instead of CS. The support estimate of the modified-CS reconstruction is, in turn, used to update the correlation model parameters using a Kalman filter (or any adaptive filter). We call the resulting approach "support-predicted modified-CS".

In this work, we obtain sufficient conditions for the "stability" of our recently proposed algorithms, Least Squares Compressive Sensing residual (LS-CS) and modified-CS, for recursively reconstructing sparse signal sequences from noisy measurements. By "stability" we mean that the number of misses from the current support estimate and the number of extras in it remain bounded by a time-invariant value at all times. We show that, for a signal model with fixed signal power and support set size; support set changes allowed at every time; and gradual coefficient magnitude increase/decrease, "stability" holds under mild assumptions -- bounded noise, high enough minimum nonzero coefficient magnitude increase rate, and large enough number of measurements at every time. A direct corollary is that the reconstruction error is also bounded by a time-invariant value at all times. If the support set of the sparse signal sequence changes slowly over time, our results hold under weaker assumptions than what simple compressive sensing (CS) needs for the same error bound. Also, our support error bounds are small compared to the support size. Our discussion is backed up by Monte Carlo simulation based comparisons., Comment: 16 pages, 3 figures. Submitted to IEEE Trans. Information Theory. Shorter version in Allerton 2010 (Stability (over time) of Modified-CS for Recursive Causal Sparse Reconstruction)

This work proposes a causal and recursive algorithm for solving the "robust" principal components' analysis (PCA) problem. We primarily focus on robustness to correlated outliers. In recent work, we proposed a new way to look at this problem and showed how a key part of its solution strategy involves solving a noisy compressive sensing(CS) problem. However, if the support size of the outliers becomes too large, for a given dimension of the current PC space, then the number of "measurements" available for CS may become too small. In this work, we show how to address this issue by utilizing the correlation of the outliers to predict their support at the current time; and using this as "partial support knowledge" for solving Modified-CS instead of CS.

In the recent work of Candes et al, the problem of recovering low rank matrix corrupted by i.i.d. sparse outliers is studied and a very elegant solution, principal component pursuit, is proposed. It is motivated as a tool for video surveillance applications with the background image sequence forming the low rank part and the moving objects/persons/abnormalities forming the sparse part. Each image frame is treated as a column vector of the data matrix made up of a low rank matrix and a sparse corruption matrix. Principal component pursuit solves the problem under the assumptions that the singular vectors of the low rank matrix are spread out and the sparsity pattern of the sparse matrix is uniformly random. However, in practice, usually the sparsity pattern and the signal values of the sparse part (moving persons/objects) change in a correlated fashion over time, for e.g., the object moves slowly and/or with roughly constant velocity. This will often result in a low rank sparse matrix. For video surveillance applications, it would be much more useful to have a real-time solution. In this work, we study the online version of the above problem and propose a solution that automatically handles correlated sparse outliers. The key idea of this work is as follows. Given an initial estimate of the principal directions of the low rank part, we causally keep estimating the sparse part at each time by solving a noisy compressive sensing type problem. The principal directions of the low rank part are updated every-so-often. In between two update times, if new Principal Components' directions appear, the "noise" seen by the Compressive Sensing step may increase. This problem is solved, in part, by utilizing the time correlation model of the low rank part. We call the proposed solution "Real-time Robust Principal Components' Pursuit"., Comment: 8 pages, 4 figures, Allerton 2010

In this work, we obtain sufficient conditions for the ``stability" of our recently proposed algorithms, modified-CS (for noisy measurements) and Least Squares CS-residual (LS-CS), designed for recursive reconstruction of sparse signal sequences from noisy measurements. By ``stability" we mean that the number of misses from the current support estimate and the number of extras in it remain bounded by a time-invariant value at all times. The concept is meaningful only if the bound is small compared to the current signal support size. A direct corollary is that the reconstruction errors are also bounded by a time-invariant and small value., Comment: 15 pages, one figure with four rows

We study the problem of sparse reconstruction from noisy undersampled measurements when the following two things are available. (1) We are given partial, and partly erroneous, knowledge of the signal's support, denoted by $T$. (2) We are also given an erroneous estimate of the signal values on $T$, denoted by $(\hat{\mu})_T$. In practice, both these may be available from available prior knowledge. Alternatively, in recursive reconstruction applications, like real-time dynamic MRI, one can use the support estimate and the signal value estimate from the previous time instant as $T$ and $(\hat{\mu})_T$. In this work, we introduce regularized modified-BPDN (reg-mod-BPDN) and obtain computable bounds on its reconstruction error. Reg-mod-BPDN tries to find the signal that is sparsest outside the set $T$, while being "close enough" to $(\hat{\mu})_T$ on $T$ and while satisfying the data constraint. Corresponding results for modified-BPDN and BPDN follow as direct corollaries. A second key contribution is an approach to obtain computable error bounds that hold without any sufficient conditions. This makes it easy to compare the bounds for the various approaches. Empirical reconstruction error comparisons with many existing approaches are also provided., Comment: 30 pages, 5 figures

We consider the problem of recursively reconstructing time sequences of sparse signals (with unknown and time-varying sparsity patterns) from a limited number of linear incoherent measurements with additive noise. The idea of our proposed solution, KF CS-residual (KF-CS) is to replace compressed sensing (CS) on the observation by CS on the Kalman filtered (KF) observation residual computed using the previous estimate of the support. KF-CS error stability over time is studied. Simulation comparisons with CS and LS-CS are shown., Comment: 7 pages, 2 figures, submitted to the IEEE for possible publication

We consider the problem of recursively and causally reconstructing time sequences of sparse signals (with unknown and time-varying sparsity patterns) from a limited number of noisy linear measurements. The sparsity pattern is assumed to change slowly with time. The idea of our proposed solution, LS-CS-residual (LS-CS), is to replace compressed sensing (CS) on the observation by CS on the least squares (LS) residual computed using the previous estimate of the support. We bound CS-residual error and show that when the number of available measurements is small, the bound is much smaller than that on CS error if the sparsity pattern changes slowly enough. We also obtain conditions for "stability" of LS-CS over time for a signal model that allows support additions and removals, and that allows coefficients to gradually increase (decrease) until they reach a constant value (become zero). By "stability", we mean that the number of misses and extras in the support estimate remain bounded by time-invariant values (in turn implying a time-invariant bound on LS-CS error). The concept is meaningful only if the bounds are small compared to the support size. Numerical experiments backing our claims are shown., Comment: Accepted (with mandatory minor revisions) to IEEE Trans. Signal Processing. 12 pages, 5 figures

In this work, we study the problem of reconstructing a sparse signal from a limited number of linear 'incoherent' noisy measurements, when a part of its support is known. The known part of the support may be available from prior knowledge or from the previous time instant (in applications requiring recursive reconstruction of a time sequence of sparse signals, e.g. dynamic MRI). We study a modification of Basis Pursuit Denoising (BPDN) and bound its reconstruction error. A key feature of our work is that the bounds that we obtain are computable. Hence, we are able to use Monte Carlo to study their average behavior as the size of the unknown support increases. We also demonstrate that when the unknown support size is small, modified-BPDN bounds are much tighter than those for BPDN, and hold under much weaker sufficient conditions (require fewer measurements)., Comment: 5 pages, 3 figures, submitted ICASSP 2010