In recent years, stochastic gradient descent (SGD) methods and randomized linear algebra (RLA) algorithms have been applied to many large-scale problems in machine learning and data analysis. We aim to bridge the gap between these two methods in solving constrained overdetermined linear regression problems---e.g., $\ell_2$ and $\ell_1$ regression problems. We propose a hybrid algorithm named pwSGD that uses RLA techniques for preconditioning and constructing an importance sampling distribution, and then performs an SGD-like iterative process with weighted sampling on the preconditioned system. We prove that pwSGD inherits faster convergence rates that only depend on the lower dimension of the linear system, while maintaining low computation complexity. Particularly, when solving $\ell_1$ regression with size $n$ by $d$, pwSGD returns an approximate solution with $ε$ relative error in the objective value in $\mathcal{O}(\log n \cdot \text{nnz}(A) + \text{poly}(d)/ε^2)$ time. This complexity is uniformly better than that of RLA methods in terms of both $ε$ and $d$ when the problem is unconstrained. For $\ell_2$ regression, pwSGD returns an approximate solution with $ε$ relative error in the objective value and the solution vector measured in prediction norm in $\mathcal{O}(\log n \cdot \text{nnz}(A) + \text{poly}(d) \log(1/ε) /ε)$ time. We also provide lower bounds on the coreset complexity for more general regression problems, indicating that still new ideas will be needed to extend similar RLA preconditioning ideas to weighted SGD algorithms for more general regression problems. Finally, the effectiveness of such algorithms is illustrated numerically on both synthetic and real datasets.
Technical University Berlin（柏林工业大学）；University of Potsdam（波茨坦大学）
Stochastic Gradient Descent (SGD) has become the method of choice for solving a broad range of machine learning problems. However, some of its learning properties are still not fully understood. We consider least squares learning in reproducing kernel Hilbert spaces (RKHSs) and extend the classical SGD analysis to a learning setting in Hilbert scales, including Sobolev spaces and Diffusion spaces on compact Riemannian manifolds. We show that even for well-specified models, violation of a traditional benchmark smoothness assumption has a tremendous effect on the learning rate. In addition, we s…
Mila；McGill University（麦吉尔大学）；Stanford University（斯坦福大学）；École Polytechnique de Montréal（蒙特利尔理工学院）；Université de Montréal（蒙特利尔大学）；Facebook AI Research
We investigate whether Jacobi preconditioning, accounting for the bootstrap term in temporal difference (TD) learning, can help boost performance of adaptive optimizers. Our method, TDprop, computes a per parameter learning rate based on the diagonal preconditioning of the TD update rule. We show how this can be used in both $n$-step returns and TD($λ$). Our theoretical findings demonstrate that including this additional preconditioning information is, surprisingly, comparable to normal semi-gradient TD if the optimal learning rate is found for both via a hyperparameter search. In Deep RL expe…
Cyclic and Randomized Stepsizes Invoke Heavier Tails in SGD than Constant Stepsize
Rutgers Business School（罗格斯商学院）；Princeton University（普林斯顿大学）；Inria（法国国家信息与自动化研究所）；CNRS（法国国家科学研究中心）；Ecole Normale Supérieure（巴黎高等师范学院）；PSL Research University（巴黎文理研究大学）；Florida State University（佛罗里达州立大学）
An increasing number of machine learning problems, such as robust or adversarial variants of existing algorithms, require minimizing a loss function that is itself defined as a maximum. Carrying a loop of stochastic gradient ascent (SGA) steps on the (inner) maximization problem, followed by an SGD step on the (outer) minimization, is known as Epoch Stochastic Gradient \textit{Descent Ascent} (ESGDA). While successful in practice, the theoretical analysis of ESGDA remains challenging, with no clear guidance on choices for the inner loop size nor on the interplay between inner/outer step sizes.…
Preconditioned gradient methods are among the most general and powerful tools in optimization. However, preconditioning requires storing and manipulating prohibitively large matrices. We describe and analyze a new structure-aware preconditioning algorithm, called Shampoo, for stochastic optimization over tensor spaces. Shampoo maintains a set of preconditioning matrices, each of which operates on a single dimension, contracting over the remaining dimensions. We establish convergence guarantees in the stochastic convex setting, the proof of which builds upon matrix trace inequalities. Our exper…
Duke University（杜克大学）；University of Science and Technology of China（中国科学技术大学）；Microsoft Research Asia（微软亚洲研究院）；The Ohio State University（俄亥俄州立大学）
Stochastic gradient descent (SGD) has been found to be surprisingly effective in training a variety of deep neural networks. However, there is still a lack of understanding on how and why SGD can train these complex networks towards a global minimum. In this study, we establish the convergence of SGD to a global minimum for nonconvex optimization problems that are commonly encountered in neural network training. Our argument exploits the following two important properties: 1) the training loss can achieve zero value (approximately), which has been widely observed in deep learning; 2) SGD follo…
Moscow Institute of Physics and Technology（莫斯科物理技术学院）；King Abdullah University of Science and Technology (KAUST)（阿卜杜拉国王科技大学）
Most algorithms for solving optimization problems or finding saddle points of convex-concave functions are fixed-point algorithms. In this work we consider the generic problem of finding a fixed point of an average of operators, or an approximation thereof, in a distributed setting. Our work is motivated by the needs of federated learning. In this context, each local operator models the computations done locally on a mobile device. We investigate two strategies to achieve such a consensus: one based on a fixed number of local steps, and the other based on randomized computations. In both cases…
We show that Entropy-SGD (Chaudhari et al., 2017), when viewed as a learning algorithm, optimizes a PAC-Bayes bound on the risk of a Gibbs (posterior) classifier, i.e., a randomized classifier obtained by a risk-sensitive perturbation of the weights of a learned classifier. Entropy-SGD works by optimizing the bound's prior, violating the hypothesis of the PAC-Bayes theorem that the prior is chosen independently of the data. Indeed, available implementations of Entropy-SGD rapidly obtain zero training error on random labels and the same holds of the Gibbs posterior. In order to obtain a valid g…
Quebec Artificial Intelligence Institute (Mila)（魁北克人工智能研究所（米拉））；Université de Montréal（蒙特利尔大学）；University of Wisconsin-Madison（威斯康星大学麦迪逊分校）；University of Athens（雅典大学）
Several works have aimed to explain why overparameterized neural networks generalize well when trained by Stochastic Gradient Descent (SGD). The consensus explanation that has emerged credits the randomized nature of SGD for the bias of the training process towards low-complexity models and, thus, for implicit regularization. We take a careful look at this explanation in the context of image classification with common deep neural network architectures. We find that if we do not regularize \emph{explicitly}, then SGD can be easily made to converge to poorly-generalizing, high-complexity models:…
Universitat Pompeu Fabra（庞培法布拉大学）；Element AI（Element AI公司）；Mila（米拉实验室）；University of Toronto（多伦多大学）；Vector Institute（矢量研究院）
We study the generalization properties of the popular stochastic optimization method known as stochastic gradient descent (SGD) for optimizing general non-convex loss functions. Our main contribution is providing upper bounds on the generalization error that depend on local statistics of the stochastic gradients evaluated along the path of iterates calculated by SGD. The key factors our bounds depend on are the variance of the gradients (with respect to the data distribution) and the local smoothness of the objective function along the SGD path, and the sensitivity of the loss function to pert…
SGD for Variational Inference: Tackling Unbounded Variance via Preconditioning and Dynamic Batching
MaLGa；DIBRIS Università degli Studi di Genova（DIBRIS基因瓦大学）；Università degli Studi di Genova（基因瓦大学）；DIMA Università degli Studi di Genova（DIMA基因瓦大学）；IIT Università degli Studi di Genova（IIT基因瓦大学）
Asymptotic Analysis of Conditioned Stochastic Gradient Descent
École Polytechnique（巴黎综合理工学院）；Institut Polytechnique de Paris（巴黎理工学院）；ENSAI（法国国立统计与信息分析学院）；École Nationale de la Statistique et de l’Analyse de l’Information（法国国立统计与信息分析学院）
We establish novel generalization bounds for learning algorithms that converge to global minima. We do so by deriving black-box stability results that only depend on the convergence of a learning algorithm and the geometry around the minimizers of the loss function. The results are shown for nonconvex loss functions satisfying the Polyak-Łojasiewicz (PL) and the quadratic growth (QG) conditions. We further show that these conditions arise for some neural networks with linear activations. We use our black-box results to establish the stability of optimization algorithms such as stochastic gradi…
Exponential generalization bounds with near-tight rates have recently been established for uniformly stable learning algorithms. The notion of uniform stability, however, is stringent in the sense that it is invariant to the data-generating distribution. Under the weaker and distribution dependent notions of stability such as hypothesis stability and $L_2$-stability, the literature suggests that only polynomial generalization bounds are possible in general cases. The present paper addresses this long standing tension between these two regimes of results and makes progress towards relaxing it i…
Interior point methods solve small to medium sized problems to high accuracy in a reasonable amount of time. However, for larger problems as well as stochastic problems, one needs to use first-order methods such as stochastic gradient descent (SGD), the alternating direction method of multipliers (ADMM), and conjugate gradient (CG) in order to attain a modest accuracy in a reasonable number of iterations. In this report, we first discuss heuristics for diagonal scaling. Next, we motivate preconditioning by an example, and then we study preconditioning for a specific splitting form in ADMM call…
Momentum Iterative Hessian Sketch (M-IHS) techniques, a group of solvers for large scale regularized linear Least Squares (LS) problems, are proposed and analyzed in detail. Proposed M-IHS techniques are obtained by incorporating the Heavy Ball Acceleration into the Iterative Hessian Sketch algorithm and they provide significant improvements over the randomized preconditioning techniques. By using approximate solvers along with the iterations, the proposed techniques are capable of avoiding all matrix decompositions and inversions, which is one of the main advantages over the alternative solve…
Department of Computer Science and Engineering（计算机科学与工程系）；State University of New York at Buffalo（纽约州立大学布法罗分校）
In this paper, we revisit the large-scale constrained linear regression problem and propose faster methods based on some recent developments in sketching and optimization. Our algorithms combine (accelerated) mini-batch SGD with a new method called two-step preconditioning to achieve an approximate solution with a time complexity lower than that of the state-of-the-art techniques for the low precision case. Our idea can also be extended to the high precision case, which gives an alternative implementation to the Iterative Hessian Sketch (IHS) method with significantly improved time complexity.…
The paper studies the properties of stochastic gradient methods with preconditioning. We focus on momentum updated preconditioners with momentum coefficient $β$. Seeking to explain practical efficiency of scaled methods, we provide convergence analysis in a norm associated with preconditioner, and demonstrate that scaling allows one to get rid of gradients Lipschitz constant in convergence rates. Along the way, we emphasize important role of $β$, undeservedly set to constant $0.99...9$ at the arbitrariness of various authors. Finally, we propose the explicit constructive formulas for adaptive …
DeepMind；Google（谷歌）；Google Research - Brain（谷歌研究院 - Brain）
Differentially Private Stochastic Gradient Descent (DP-SGD) forms a fundamental building block in many applications for learning over sensitive data. Two standard approaches, privacy amplification by subsampling, and privacy amplification by shuffling, permit adding lower noise in DP-SGD than via na\"ıve schemes. A key assumption in both these approaches is that the elements in the data set can be uniformly sampled, or be uniformly permuted -- constraints that may become prohibitive when the data is processed in a decentralized or distributed fashion. In this paper, we focus on conducting iter…
School of Mathematics（数学学院）；University of Edinburgh（爱丁堡大学）
In this paper we consider the problem of minimizing a convex function using a randomized block coordinate descent method. One of the key steps at each iteration of the algorithm is determining the update to a block of variables. Existing algorithms assume that in order to compute the update, a particular subproblem is solved exactly. In his work we relax this requirement, and allow for the subproblem to be solved inexactly, leading to an inexact block coordinate descent method. Our approach incorporates the best known results for exact updates as a special case. Moreover, these theoretical gua…
University of California, Berkeley（加州大学伯克利分校）；Stanford University（斯坦福大学）
Variance reduction methods such as SVRG and SpiderBoost use a mixture of large and small batch gradients to reduce the variance of stochastic gradients. Compared to SGD, these methods require at least double the number of operations per update to model parameters. To reduce the computational cost of these methods, we introduce a new sparsity operator: The random-top-k operator. Our operator reduces computational complexity by estimating gradient sparsity exhibited in a variety of applications by combining the top-k operator and the randomized coordinate descent operator. With this operator, la…
This work presents stochastic optimization methods targeted at least-squares problems involving Monte Carlo integration. While the most common approach to solving these problems is to apply stochastic gradient descent (SGD) or similar methods such as AdaGrad and Adam, which involve estimating a stochastic gradient from a small number of Monte Carlo samples computed at each iteration, we show that for this category of problems it is possible to achieve faster asymptotic convergence rates using an increasing number of samples per iteration instead, a strategy we call increasing precision (IP). W…
KTH Royal Institute of Technology（瑞典皇家理工学院）；Columbia University（哥伦比亚大学）；Princeton University（普林斯顿大学）；Institute for Advanced Study（高等研究院）
The slow convergence rate and pathological curvature issues of first-order gradient methods for training deep neural networks, initiated an ongoing effort for developing faster $\mathit{second}$-$\mathit{order}$ optimization algorithms beyond SGD, without compromising the generalization error. Despite their remarkable convergence rate ($\mathit{independent}$ of the training batch size $n$), second-order algorithms incur a daunting slowdown in the $\mathit{cost}$ $\mathit{per}$ $\mathit{iteration}$ (inverting the Hessian matrix of the loss function), which renders them impractical. Very recentl…
University of Bath（巴斯大学）；University of Cambridge（剑桥大学）
Uncompressed clinical data from modern positron emission tomography (PET) scanners are very large, exceeding 350 million data points (projection bins). The last decades have seen tremendous advancements in mathematical imaging tools many of which lead to non-smooth (i.e. non-differentiable) optimization problems which are much harder to solve than smooth optimization problems. Most of these tools have not been translated to clinical PET data, as the state-of-the-art algorithms for non-smooth problems do not scale well to large data. In this work, inspired by big data machine learning applicati…
Toyota Technological Institute at Chicago（芝加哥丰田技术学院）
In this thesis, I study the minimax oracle complexity of distributed stochastic optimization. First, I present the "graph oracle model", an extension of the classic oracle complexity framework that can be applied to study distributed optimization algorithms. Next, I describe a general approach to proving optimization lower bounds for arbitrary randomized algorithms (as opposed to more restricted classes of algorithms, e.g., deterministic or "zero-respecting" algorithms), which is used extensively throughout the thesis. For the remainder of the thesis, I focus on the specific case of the "inter…
