Nonconvex optimization is central in solving many machine learning problems, in which block-wise structure is commonly encountered. In this work, we propose cyclic block coordinate methods for nonconvex optimization problems with non-asymptotic gradient norm guarantees. Our convergence analysis is based on a gradient Lipschitz condition with respect to a Mahalanobis norm, inspired by a recent progress on cyclic block coordinate methods. In deterministic settings, our convergence guarantee matches the guarantee of (full-gradient) gradient descent, but with the gradient Lipschitz constant being defined w.r.t. a Mahalanobis norm. In stochastic settings, we use recursive variance reduction to decrease the per-iteration cost and match the arithmetic operation complexity of current optimal stochastic full-gradient methods, with a unified analysis for both finite-sum and infinite-sum cases. We prove a faster linear convergence result when a Polyak-Łojasiewicz (PŁ) condition holds. To our knowledge, this work is the first to provide non-asymptotic convergence guarantees — variance-reduced or not — for a cyclic block coordinate method in general composite (smooth + nonsmooth) nonconvex settings. Our experimental results demonstrate the efficacy of the proposed cyclic scheme in training deep neural nets.

Exploiting partial first-order information in a cyclic way is arguably the most natural strategy to obtain scalable first-order methods. However, despite their wide use in practice, cyclic schemes are far less understood from a theoretical perspective than their randomized counterparts. Motivated by a recent success in analyzing an extrapolated cyclic scheme for generalized variational inequalities, we propose an *Accelerated Cyclic Coordinate Dual Averaging with Extrapolation* (A-CODER) method for composite convex optimization, where the objective function can be expressed as the sum of a smooth convex function accessible via a gradient oracle and a convex, possibly nonsmooth, function accessible via a proximal oracle. We show that A-CODER attains the optimal convergence rate with improved dependence on the number of blocks compared to prior work. Furthermore, for the setting where the smooth component of the objective function is expressible in a finite sum form, we introduce a variance-reduced variant of A-CODER, VR-A-CODER, with state-of-the-art complexity guarantees. Finally, we demonstrate the effectiveness of our algorithms through numerical experiments.

We study stochastic monotone inclusion problems, which widely appear in machine learning applications, including robust regression and adversarial learning. We propose novel variants of stochastic Halpern iteration with recursive variance reduction. In the cocoercive---and more generally Lipschitz-monotone---setup, our algorithm attains $\epsilon$ norm of the operator with $\mathcal{O}(\frac{1}{\epsilon^3})$ stochastic operator evaluations, which significantly improves over state of the art $\mathcal{O}(\frac{1}{\epsilon^4})$ stochastic operator evaluations required for existing monotone inclusion solvers applied to the same problem classes. We further show how to couple one of the proposed variants of stochastic Halpern iteration with a scheduled restart scheme to solve stochastic monotone inclusion problems with ${\mathcal{O}}(\frac{\log(1/\epsilon)}{\epsilon^2})$ stochastic operator evaluations under additional sharpness or strong monotonicity assumptions.

Nonnegative (linear) least square problems are a fundamental class of problems that is well-studied in statistical learning and for which solvers have been implemented in many of the standard programming languages used within the machine learning community. The existing off-the-shelf solvers view the non-negativity constraint in these problems as an obstacle and, compared to unconstrained least squares, perform additional effort to address it. However, in many of the typical applications, the data itself is nonnegative as well, and we show that the nonnegativity in this case makes the problem easier. In particular, while the worst-case dimension-independent oracle complexity of unconstrained least squares problems necessarily scales with one of the data matrix constants (typically the spectral norm) and these problems are solved to additive error, we show that nonnegative least squares problems with nonnegative data are solvable to multiplicative error and with complexity that is independent of any matrix constants. The algorithm we introduce is accelerated and based on a primal-dual perspective. We further show how to provably obtain linear convergence using adaptive restart coupled with our method and demonstrate its effectiveness on large-scale data via numerical experiments.

We study a class of generalized linear programs (GLP) in a large-scale setting, which includes simple, possibly nonsmooth convex regularizer and simple convex set constraints. By reformulating (GLP) as an equivalent convex-concave min-max problem, we show that the linear structure in the problem can be used to design an efficient, scalable first-order algorithm, to which we give the name Coordinate Linear Variance Reduction (CLVR; pronounced ``clever''). CLVR yields improved complexity results for (GLP) that depend on the max row norm of the linear constraint matrix in (GLP) rather than the spectral norm. When the regularization terms and constraints are separable, CLVR admits an efficient lazy update strategy that makes its complexity bounds scale with the number of nonzero elements of the linear constraint matrix in (GLP) rather than the matrix dimensions. On the other hand, for the special case of linear programs, by exploiting sharpness, we propose a restart scheme for CLVR to obtain empirical linear convergence. Then we show that Distributionally Robust Optimization (DRO) problems with ambiguity sets based on both $f$-divergence and Wasserstein metrics can be reformulated as (GLPs) by introducing sparsely connected auxiliary variables. We complement our theoretical guarantees with numerical experiments that verify our algorithm's practical effectiveness, in terms of wall-clock time and number of data passes.

Structured nonsmooth convex finite-sum optimization appears in many machine learning applications, including support vector machines and least absolute deviation. For the primal-dual formulation of this problem, we propose a novel algorithm called \emph{Variance Reduction via Primal-Dual Accelerated Dual Averaging (\vrpda)}. In the nonsmooth and general convex setting, \vrpda has the overall complexity $O(nd\log\min \{1/\epsilon, n\} + d/\epsilon )$ in terms of the primal-dual gap, where $n$ denotes the number of samples, $d$ the dimension of the primal variables, and $\epsilon$ the desired accuracy. In the nonsmooth and strongly convex setting, the overall complexity of \vrpda becomes $O(nd\log\min\{1/\epsilon, n\} + d/\sqrt{\epsilon})$ in terms of both the primal-dual gap and the distance between iterate and optimal solution. Both these results for \vrpda improve significantly on state-of-the-art complexity estimates—which are $O(nd\log \min\{1/\epsilon, n\} + \sqrt{n}d/\epsilon)$ for the nonsmooth and general convex setting and $O(nd\log \min\{1/\epsilon, n\} + \sqrt{n}d/\sqrt{\epsilon})$ for the nonsmooth and strongly convex setting—with a simpler and more straightforward algorithm and analysis. Moreover, both complexities are better than \emph{lower} bounds for general convex finite-sum optimization, because our approach makes use of additional, commonly occurring structure. Numerical experiments reveal competitive performance of \vrpda compared to state-of-the-art approaches.

In this paper, we introduce a simplified and unified method for finite-sum convex optimization, named \emph{Variance Reduction via Accelerated Dual Averaging (VRADA)}. In the general convex and smooth setting, VRADA can attain an $O\big(\frac{1}{n}\big)$-accurate solution in $O(n\log\log n)$ number of stochastic gradient evaluations, where $n$ is the number of samples; meanwhile, VRADA matches the lower bound of this setting up to a $\log\log n$ factor. In the strongly convex and smooth setting, VRADA matches the lower bound in the regime $n \le \Theta(\kappa)$, while it improves the rate in the regime $n\gg \kappa$ to $O\big(n +\frac{n\log(1/\epsilon)}{\log(n/\kappa)}\big)$, where $\kappa$ is the condition number. Besides improving the best known complexity results, VRADA has more unified and simplified algorithmic implementation and convergence analysis for both the general convex and strongly convex settings. Through experiments on real datasets, we show the good performance of VRADA over existing methods for large-scale machine learning problems.

The optimization problems associated with training generative adversarial neural networks can be largely reduced to certain {\em non-monotone} variational inequality problems (VIPs), whereas existing convergence results are mostly based on monotone or strongly monotone assumptions. In this paper, we propose {\em optimistic dual extrapolation (OptDE)}, a method that only performs {\em one} gradient evaluation per iteration. We show that OptDE is provably convergent to {\em a strong solution} under different coherent non-monotone assumptions. In particular, when a {\em weak solution} exists, the convergence rate of our method is $O(1/{\epsilon^{2}})$, which matches the best existing result of the methods with two gradient evaluations. Further, when a {\em $\sigma$-weak solution} exists, the convergence guarantee is improved to the linear rate $O(\log\frac{1}{\epsilon})$. Along the way--as a byproduct of our inquiries into non-monotone variational inequalities--we provide the near-optimal $O\big(\frac{1}{\epsilon}\log \frac{1}{\epsilon}\big)$ convergence guarantee in terms of restricted strong merit function for monotone variational inequalities. We also show how our results can be naturally generalized to the stochastic setting, and obtain corresponding new convergence results. Taken together, our results contribute to the broad landscape of variational inequality--both non-monotone and monotone alike--by providing a novel and more practical algorithm with the state-of-the-art convergence guarantees.

To learn intrinsic low-dimensional structures from high-dimensional data that most discriminate between classes, we propose the principle of {\em Maximal Coding Rate Reduction} ($\text{MCR}^2$), an information-theoretic measure that maximizes the coding rate difference between the whole dataset and the sum of each individual class. We clarify its relationships with most existing frameworks such as cross-entropy, information bottleneck, information gain, contractive and contrastive learning, and provide theoretical guarantees for learning diverse and discriminative features. The coding rate can be accurately computed from finite samples of degenerate subspace-like distributions and can learn intrinsic representations in supervised, self-supervised, and unsupervised settings in a unified manner. Empirically, the representations learned using this principle alone are significantly more robust to label corruptions in classification than those using cross-entropy, and can lead to state-of-the-art results in clustering mixed data from self-learned invariant features.

Optimal Transport (OT) is dedicated to solving how to transform one measure to another with least cost. OT has gained wide applications in machine learning. However, the heavy computation burden of primal OT distance makes it prohibitive for prevalent high-dimensional problems. Recent work imposes an entropic regularization term on primal OT, obtaining a strictly convex problem that can be addressed with stochastic optimization. We focus on the optimization methods for discrete OT and semi-discrete OT with regularization. Instead of the initial SAG method for discrete OT, we apply SAGA, which solves the biased gradient estimation problem, and SVRG, which solves the memory problem. Besides, we define the regret in semi-discrete OT problem and solve it from the perspective of online learning. Our study, to our best knowledge, is the first effort to solve the semi-discrete OT problem with follow-the-regularized-leader thought. Our FTRL-\textit{current} algorithm shows faster convergence than existing algorithms for solving semi-discrete OT problem.

In this paper we study the well-known greedy coordinate descent (GCD) algorithm to solve $\ell_1$-regularized problems and improve GCD by the two popular strategies: Nesterov's acceleration and stochastic optimization. Firstly, we propose a new rule for greedy selection based on an $\ell_1$-norm square approximation which is nontrivial to solve but convex; then an efficient algorithm called ``SOft ThreshOlding PrOjection (SOTOPO)'' is proposed to exactly solve the $\ell_1$-regularized $\ell_1$-norm square approximation problem, which is induced by the new rule. Based on the new rule and the SOTOPO algorithm, the Nesterov's acceleration and stochastic optimization strategies are then successfully applied to the GCD algorithm. The resulted algorithm called accelerated stochastic greedy coordinate descent (ASGCD) has the optimal convergence rate $O(\sqrt{1/\epsilon})$; meanwhile, it reduces the iteration complexity of greedy selection up to a factor of sample size. Both theoretically and empirically, we show that ASGCD has better performance for high-dimensional and dense problems with sparse solution.

The construction of efficient and effective decision trees remains a key topic in machine learning because of their simplicity and flexibility. A lot of heuristic algorithms have been proposed to construct near-optimal decision trees. Most of them, however, are greedy algorithms that have the drawback of obtaining only local optimums. Besides, conventional split criteria they used, e.g. Shannon entropy, Gain Ratio and Gini index, cannot select informative attributes efficiently. To address the above issues, we propose a novel Tsallis Entropy Information Metric (TEIM) algorithm with a new split criterion and a new construction method of decision trees. Firstly, the new split criterion is based on two terms of Tsallis conditional entropy, which is better than conventional split criteria. Secondly, the new construction method is based on a two-stage approach that avoids local optimum to a certain extent. The TEIM algorithm takes advantages of the generalization ability of Tsallis entropy and the low greediness property of two-stage approach. Experimental results on UCI datasets indicate that, compared with the state-of-the-art decision trees algorithms, the TEIM algorithm yields statistically significantly better decision trees in classification accuracy as well as tree complexity.

Patch-based algorithms for image denoising have been widely used in recent years. Most of patch-based methods just exploit patch redundancy in spatial or frequency domain without considering inter-scale dependencies. In this paper, we propose a novel patch-based multiscale products algorithm (PMPA) for image denoising. It is based on patch similarity in spatial domain and multiscale products in wavelet domain. PMPA is divided into two stages to process the smooth areas and non smooth areas (such as edges) individually. The first stage is in the wavelet domain, then a locally adaptive window-based denoising method (LAWML) based on multiscale products is applied to process those wavelet coefficients corresponding to the non smooth areas, then obtain one initial denoised image. The second stage is in the spatial domain, then a non local means algorithm is used to process those pixels in the smooth areas to obtain another initial denoised image. The final denoised image is obtained by a weighted averaging of all common pixels in both initial denoised images. Experiments show that the proposed algorithm can have competitive performance compared with the state-of-the-art patch-based denoising algorithms for most of images.

We propose a new iterative greedy algorithm to reconstruct sparse signals in compressed sensing. By a simple combination of subspace pursuit and iterative hard thresholding, the proposed algorithm, called subspace thresholding pursuit (STP) shows well improved empirical performance, while still keeps a strong theoretical guarantee in terms of restricted isometry property.

In the context of compressed sensing (CS), subspace pursuit (SP) is an important iterative greedy recovery algorithm which could reduce the recovery complexity greatly comparing with${\ell _1}$-minimization. Restricted isometry property (RIP) and restricted isometry constant (RIC) of measurement matrices which ensure the convergence of iterative algorithms play key roles for the guarantee of successful reconstructions. In this letter, we show that for the$s$-sparse recovery, the RIC is enlarged to${\delta _{3s}} < 0.4859$for SP, which improves the known results significantly. The proposed result also applies to almost sparse signals and corrupted measurements.