In this work, we develop first-order (Hessian-free) and zero-order (derivative-free) implementations of the Cubically regularized Newton method for solving general non-convex optimization problems. For that, we employ finite difference approximations of the derivatives. We use a special adaptive search procedure in our algorithms, which simultaneously fits both the regularization constant and the parameters of the finite difference approximations. It makes our schemes free from the need to know the actual Lipschitz constants. Additionally, we equip our algorithms with the lazy Hessian update that reuse a previously computed Hessian approximation matrix for several iterations. Specifically, we prove the global complexity bound of $\mathcal{O}( n^{1/2} ε^{-3/2})$ function and gradient evaluations for our new Hessian-free method, and a bound of $\mathcal{O}( n^{3/2} ε^{-3/2} )$ function evaluations for the derivative-free method, where $n$ is the dimension of the problem and $ε$ is the desired accuracy for the gradient norm. These complexity bounds significantly improve the previously known ones in terms of the joint dependence on $n$ and $ε$, for the first-order and zeroth-order non-convex optimization.
A major challenge of applying zeroth-order (ZO) methods is the high query complexity, especially when queries are costly. We propose a novel gradient estimation technique for ZO methods based on adaptive lazy queries that we term as LAZO. Different from the classic one-point or two-point gradient estimation methods, LAZO develops two alternative ways to check the usefulness of old queries from previous iterations, and then adaptively reuses them to construct the low-variance gradient estimates. We rigorously establish that through judiciously reusing the old queries, LAZO can reduce the varian…
Stochastic Zeroth-Order Optimization under Strongly Convexity and Lipschitz Hessian: Minimax Sample Complexity
University of California, Santa Barbara（加利福尼亚大学圣巴巴拉分校）；University of Texas at Dallas（德克萨斯大学达拉斯分校）；University of California, Berkeley（加利福尼亚大学伯克利分校）；New York University（纽约大学）；Princeton University（普林斯顿大学）
Nanjing University of Information Science & Technology（南京信息工程大学）；Harbin Institute of Technology（哈尔滨工业大学）；Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
We consider escaping saddle points of nonconvex problems where only the function evaluations can be accessed. Although a variety of works have been proposed, the majority of them require either second or first-order information, and only a few of them have exploited zeroth-order methods, particularly the technique of negative curvature finding with zeroth-order methods which has been proven to be the most efficient method for escaping saddle points. To fill this gap, in this paper, we propose two zeroth-order negative curvature finding frameworks that can replace Hessian-vector product computa…
The Ohio State University（俄亥俄州立大学）；University of Michigan, Ann Arbor（密歇根大学安娜堡分校）
Bilevel optimization has arisen as a powerful tool in modern machine learning. However, due to the nested structure of bilevel optimization, even gradient-based methods require second-order derivative approximations via Jacobian- or/and Hessian-vector computations, which can be costly and unscalable in practice. Recently, Hessian-free bilevel schemes have been proposed to resolve this issue, where the general idea is to use zeroth- or first-order methods to approximate the full hypergradient of the bilevel problem. However, we empirically observe that such approximation can lead to large varia…
Department of Statistics, University of California, Davis（加利福尼亚大学戴维斯分校统计系）；Department of Operations Research and Financial Engineering, Princeton University（普林斯顿大学运筹学与金融工程系）
In this paper, we propose and analyze zeroth-order stochastic approximation algorithms for nonconvex and convex optimization, with a focus on addressing constrained optimization, high-dimensional setting and saddle-point avoiding. To handle constrained optimization, we first propose generalizations of the conditional gradient algorithm achieving rates similar to the standard stochastic gradient algorithm using only zeroth-order information. To facilitate zeroth-order optimization in high-dimensions, we explore the advantages of structural sparsity assumptions. Specifically, (i) we highlight an…
Institut Polytechnique de Paris（巴黎理工学院）；JD Explore Academy（京东探索研究院）；King Abdullah University of Science and Technology（阿卜杜拉国王科技大学）
Despite their high computation and communication costs, Newton-type methods remain an appealing option for distributed training due to their robustness against ill-conditioned convex problems. In this work, we study ommunication compression and aggregation mechanisms for curvature information in order to reduce these costs while preserving theoretically superior local convergence guarantees. We prove that the recently developed class of three point compressors (3PC) of Richtarik et al. [2022] for gradient communication can be generalized to Hessian communication as well. This result opens up a…
Department of Computer Science, The Ohio State University（俄亥俄州立大学计算机科学系）；Halicioğlu Data Science Institute, University of California, San Diego（加利福尼亚大学圣迭戈分校哈利西奥卢数据科学学院）
The goal of this work is to shed light on the remarkable phenomenon of transition to linearity of certain neural networks as their width approaches infinity. We show that the transition to linearity of the model and, equivalently, constancy of the (neural) tangent kernel (NTK) result from the scaling properties of the norm of the Hessian matrix of the network as a function of the network width. We present a general framework for understanding the constancy of the tangent kernel via Hessian scaling applicable to the standard classes of neural networks. Our analysis provides a new perspective on…
Zeroth-order methods have become important tools for solving problems where we have access only to function evaluations. However, the zeroth-order methods only using gradient approximations are $n$ times slower than classical first-order methods for solving n-dimensional problems. To accelerate the convergence rate, this paper proposes the zeroth order randomized subspace Newton (ZO-RSN) method, which estimates projections of the gradient and Hessian by random sketching and finite differences. This allows us to compute the Newton step in a lower dimensional subspace, with small computational c…
Solving Convex-Concave Problems with $\tilde{\mathcal{O}}(ε^{-4/7})$ Second-Order Oracle Complexity
Tsinghua University（清华大学）；Shanghai Qizhi Institute（上海期智研究院）；The Chinese University of Hong Kong（香港中文大学）；Fudan University（复旦大学）；Shanghai AI Lab（上海人工智能实验室）
Department of Computer Science, ETH Zurich（苏黎世联邦理工学院计算机科学系）；Max Planck Institute for Intelligent Systems（智能系统马克斯·普朗克研究所）；Amazon（亚马逊公司）；Paul G. Allen School of Computer Science and Engineering, University of Washington（华盛顿大学保罗·G·阿伦计算机科学与工程学校）
In stochastic optimization problems using noisy zeroth-order (ZO) oracles only, the randomized counterpart of the Kiefer-Wolfowitz-type method is widely used to estimate the gradient. Existing algorithms generate randomized perturbation from a zero-mean and unit-covariance distribution. In contrast, this work considers the generalization where the perturbations have a possibly non-identity covariance constructed from the history of the ZO queries. We propose to feed the second-order approximation into the covariance matrix of the random perturbation, so it is dubbed as Hessian-aided random per…
We study a standard distributed optimization framework where $N$ networked nodes collaboratively minimize the sum of their local convex costs. The main body of existing work considers the described problem when the underling network is either static or deterministically varying, and the distributed optimization algorithm is of first or second order, i.e., it involves the local costs' gradients and possibly the local Hessians. In this paper, we consider the currently understudied but highly relevant scenarios when: 1) only noisy function values' estimates are available (no gradients nor Hessian…
We consider stochastic zeroth-order optimization over Riemannian submanifolds embedded in Euclidean space, where the task is to solve Riemannian optimization problem with only noisy objective function evaluations. Towards this, our main contribution is to propose estimators of the Riemannian gradient and Hessian from noisy objective function evaluations, based on a Riemannian version of the Gaussian smoothing technique. The proposed estimators overcome the difficulty of the non-linearity of the manifold constraint and the issues that arise in using Euclidean Gaussian smoothing techniques when …
When the available information is noisy zeroth-order (ZO) oracle, stochastic approximation methods are popular for estimating the root of the multivariate gradient equation. Inspired by the Stein's identity, this work establishes a novel Hessian approximation scheme. We compare it alongside with second-order simultaneous perturbation stochastic approximation (2SPSA). On the basis of the almost sure convergence and the same convergence rate, 2SPSA requires four ZO queries, while ours requires three ZO queries. Moreover, 2SPSA requires two statistically independent perturbations and two differen…
Stochastic Regret Guarantees for Online Zeroth- and First-Order Bilevel Optimization
Amirkabir University of Technology（阿姆斯泰尔大学）；University of Pennsylvania（宾夕法尼亚大学）；Samsung SDS Research America（三星SDS美国研究部）；University of California, Los Angeles（加州大学洛杉矶分校）
Smooth autonomous dynamical systems modeled by ordinary differential equations (ODEs) cannot robustly and globally stabilize a point in compact, boundaryless manifolds. This obstruction, which is topological in nature, implies that traditional smooth optimization dynamics are not able to robustly solve global optimization problems in such spaces. In turn, model-free optimization algorithms, which usually inherit their stability and convergence properties from their model-based counterparts, might also suffer from similar topological obstructions. For example, this is the case in zeroth-order m…
We consider unconstrained randomized optimization of smooth convex objective functions in the gradient-free setting. We analyze Random Pursuit (RP) algorithms with fixed (F-RP) and variable metric (V-RP). The algorithms only use zeroth-order information about the objective function and compute an approximate solution by repeated optimization over randomly chosen one-dimensional subspaces. The distribution of search directions is dictated by the chosen metric. Variable Metric RP uses novel variants of a randomized zeroth-order Hessian approximation scheme recently introduced by Leventhal and Le…
