本研究深入探讨了一类新兴的优化挑战：黑盒目标函数仅能通过排序预言机（ranking oracle）进行评估——这是现实场景中十分常见的情况，尤其是当函数由人类评判者进行评估时。该挑战的灵感来源于人类反馈强化学习（Reinforcement Learning with Human Feedback, RLHF），这是近期被用于借助人类指导提升大语言模型（Large Language Models, LLMs）性能的方法。我们提出了ZO-RankSGD，这是一种专为解决该优化问题设计的创新零阶优化算法，并附带理论保证。我们的算法采用一种新颖的基于排序的随机估计器来确定下降方向，且能够保证收敛到驻点。此外，ZO-RankSGD可直接应用于强化学习（Reinforcement Learning, RL）中的策略优化问题，特别是在仅能获得回合奖励的排序预言机的场景下。最后但同样重要的是，我们在一项新颖的应用中验证了ZO-RankSGD的有效性：利用人类排序反馈提升扩散生成模型的生成图像质量。在所有实验中我们发现，仅需少数几轮人类反馈，ZO-RankSGD就能显著提升生成图像的细节表现。总体而言，我们的工作通过解决仅具备排序反馈的函数优化问题，推动了零阶优化领域的发展，同时为人工智能（Artificial Intelligence, AI）对齐人类意图提供了一种全新且有效的方法。
英文摘要：
In this study, we delve into an emerging optimization challenge involving a black-box objective function that can only be gauged via a ranking oracle-a situation frequently encountered in real-world scenarios, especially when the function is evaluated by human judges. Such challenge is inspired from Reinforcement Learning with Human Feedback (RLHF), an approach recently employed to enhance the performance of Large Language Models (LLMs) using human guidance. We introduce ZO-RankSGD, an innovative zeroth-order optimization algorithm designed to tackle this optimization problem, accompanied by theoretical assurances. Our algorithm utilizes a novel rank-based random estimator to determine the descent direction and guarantees convergence to a stationary point. Moreover, ZO-RankSGD is readily applicable to policy optimization problems in Reinforcement Learning (RL), particularly when only ranking oracles for the episode reward are available. Last but not least, we demonstrate the effectiveness of ZO-RankSGD in a novel application: improving the quality of images generated by a diffusion generative model with human ranking feedback. Throughout experiments, we found that ZO-RankSGD can significantly enhance the detail of generated images with only a few rounds of human feedback. Overall, our work advances the field of zeroth-order optimization by addressing the problem of optimizing functions with only ranking feedback, and offers a new and effective approach for aligning Artificial Intelligence (AI) with human intentions.
Safety is a crucial necessity in many applications of reinforcement learning (RL), whether robotic, automotive, or medical. Many existing approaches to safe RL rely on receiving numeric safety feedback, but in many cases this feedback can only take binary values; that is, whether an action in a given state is safe or unsafe. This is particularly true when feedback comes from human experts. We therefore consider the problem of provable safe RL when given access to an offline oracle providing binary feedback on the safety of state, action pairs. We provide a novel meta algorithm, SABRE, which ca…
Deterministic Policy Gradient (DPG) removes a level of randomness from standard randomized-action Policy Gradient (PG), and demonstrates substantial empirical success for tackling complex dynamic problems involving Markov decision processes. At the same time, though, DPG loses its ability to learn in a model-free (i.e., actor-only) fashion, frequently necessitating the use of critics in order to obtain consistent estimates of the associated policy-reward gradient. In this work, we introduce Zeroth-order Deterministic Policy Gradient (ZDPG), which approximates policy-reward gradients via two-po…
Georgia Institute of Technology（佐治亚理工学院）；University of Pennsylvania（宾夕法尼亚大学）
Blockchains with smart contracts are distributed ledger systems that achieve block-state consistency among distributed nodes by only allowing deterministic operations of smart contracts. However, the power of smart contracts is enabled by interacting with stochastic off-chain data, which in turn opens the possibility to undermine the block-state consistency. To address this issue, an oracle smart contract is used to provide a single consistent source of external data; but, simultaneously, this introduces a single point of failure, which is called the oracle problem. To address the oracle probl…
Moscow Institute of Physics and Technology（莫斯科物理技术学院）；Higher School of Economics（高等经济学院）；Weierstrass Institute for Applied Analysis and Stochastics（魏尔斯特拉斯应用分析与随机研究所）；Institute for Information Transmission Problems RAS（俄罗斯科学院信息传输问题研究所）；Caucasus Mathematical Center, Adyghe State University（阿迪格国立大学高加索数学中心）
Saddle-point problems have recently gained increased attention from the machine learning community, mainly due to applications in training Generative Adversarial Networks using stochastic gradients. At the same time, in some applications only a zeroth-order oracle is available. In this paper, we propose several algorithms to solve stochastic smooth (strongly) convex-concave saddle-point problems using zeroth-order oracles and estimate their convergence rate and its dependence on the dimension $n$ of the variable. In particular, our analysis shows that in the case when the feasible set is a dir…
Minimisation of Quasar-Convex Functions Using Random Zeroth-Order Oracles
School of Engineering（工程学院）；Australian National University（澳大利亚国立大学）；Department of Electrical and Electronic Engineering（电子与电气工程系）；University of Melbourne（墨尔本大学）
In the learning to learn (L2L) framework, we cast the design of optimization algorithms as a machine learning problem and use deep neural networks to learn the update rules. In this paper, we extend the L2L framework to zeroth-order (ZO) optimization setting, where no explicit gradient information is available. Our learned optimizer, modeled as recurrent neural network (RNN), first approximates gradient by ZO gradient estimator and then produces parameter update utilizing the knowledge of previous iterations. To reduce high variance effect due to ZO gradient estimator, we further introduce ano…
Minimisation of Submodular Functions Using Gaussian Zeroth-Order Random Oracles
School of Engineering, Australian National University（工程学院，澳大利亚国立大学）；CIICADA Lab（CIICADA实验室）；Department of Mechanical Engineering, University of Texas at Dallas（机械工程系，德克萨斯大学达拉斯分校）；Department of Electrical and Electronic Engineering, University of Melbourne（电子与电气工程系，墨尔本大学）
Wall-Clock Complexity for Zeroth-Order Optimization with Tunable Oracle Fidelity
Weierstrass Institute for Applied Analysis and Stochastics（魏泽拉斯应用分析与随机学研究所）；Moscow Independent Research Institute of Artificial Intelligence（莫斯科独立人工智能研究 institute）；MSU Institute for Artificial Intelligence（莫斯科大学人工智能研究所）；HSE University（莫斯科国立高等经济大学）；Trusted AI Research Center, RAS（可信人工智能研究中心，俄罗斯科学院）；Innopolis University（伊诺波利斯大学）
University of California, Davis（加州大学戴维斯分校）；University of Southern California（南加州大学）
In this paper, we study zeroth-order algorithms for minimax optimization problems that are nonconvex in one variable and strongly-concave in the other variable. Such minimax optimization problems have attracted significant attention lately due to their applications in modern machine learning tasks. We first consider a deterministic version of the problem. We design and analyze the Zeroth-Order Gradient Descent Ascent (\texttt{ZO-GDA}) algorithm, and provide improved results compared to existing works, in terms of oracle complexity. We also propose the Zeroth-Order Gradient Descent Multi-Step A…
Nanjing University of Information Science & Technology（南京信息工程大学）；MBZUAI（穆罕默德·本·扎耶德人工智能大学）；JD Finance America Corporation（京东金融美国公司）；University of Pittsburgh（匹兹堡大学）
The conditional gradient algorithm (also known as the Frank-Wolfe algorithm) has recently regained popularity in the machine learning community due to its projection-free property to solve constrained problems. Although many variants of the conditional gradient algorithm have been proposed to improve performance, they depend on first-order information (gradient) to optimize. Naturally, these algorithms are unable to function properly in the field of increasingly popular zeroth-order optimization, where only zeroth-order information (function value) is available. To fill in this gap, we propose…
University of North Texas（北得克萨斯大学）；Huazhong University of Science and Technology（华中科技大学）
This paper investigates the stochastic distributed nonconvex optimization problem of minimizing a global cost function formed by the summation of $n$ local cost functions. We solve such a problem by involving zeroth-order (ZO) information exchange. In this paper, we propose a ZO distributed primal-dual coordinate method (ZODIAC) to solve the stochastic optimization problem. Agents approximate their own local stochastic ZO oracle along with coordinates with an adaptive smoothing parameter. We show that the proposed algorithm achieves the convergence rate of $\mathcal{O}(\sqrt{p}/\sqrt{T})$ for …
Gradient descent and its variants are widely used in machine learning. However, oracle access of gradient may not be available in many applications, limiting the direct use of gradient descent. This paper proposes a method of estimating gradient to perform gradient descent, that converges to a stationary point for general non-convex optimization problems. Beyond the first-order stationary properties, the second-order stationary properties are important in machine learning applications to achieve better performance. We show that the proposed model-free non-convex optimization algorithm returns …
Single-objective black box optimization (also known as zeroth-order optimization) is the process of minimizing a scalar objective $f(x)$, given evaluations at adaptively chosen inputs $x$. In this paper, we consider multi-objective optimization, where $f(x)$ outputs a vector of possibly competing objectives and the goal is to converge to the Pareto frontier. Quantitatively, we wish to maximize the standard hypervolume indicator metric, which measures the dominated hypervolume of the entire set of chosen inputs. In this paper, we introduce a novel scalarization function, which we term the hyper…
Zeroth-order optimization is the process of minimizing an objective $f(x)$, given oracle access to evaluations at adaptively chosen inputs $x$. In this paper, we present two simple yet powerful GradientLess Descent (GLD) algorithms that do not rely on an underlying gradient estimate and are numerically stable. We analyze our algorithm from a novel geometric perspective and present a novel analysis that shows convergence within an $ε$-ball of the optimum in $O(kQ\log(n)\log(R/ε))$ evaluations, for any monotone transform of a smooth and strongly convex objective with latent dimension $k < n$, wh…
Sun Yat-sen University（中山大学）；Chinese Academy of Sciences（中国科学院）
We explore whether quantum advantages can be found for the zeroth-order online convex optimization problem, which is also known as bandit convex optimization with multi-point feedback. In this setting, given access to zeroth-order oracles (that is, the loss function is accessed as a black box that returns the function value for any queried input), a player attempts to minimize a sequence of adversarially generated convex loss functions. This procedure can be described as a $T$ round iterative game between the player and the adversary. In this paper, we present quantum algorithms for the proble…
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）
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 …
This paper considers stochastic convex optimization problems with two sets of constraints: (a) deterministic constraints on the domain of the optimization variable, which are difficult to project onto; and (b) deterministic or stochastic constraints that admit efficient projection. Problems of this form arise frequently in the context of semidefinite programming as well as when various NP-hard problems are solved approximately via semidefinite relaxation. Since projection onto the first set of constraints is difficult, it becomes necessary to explore projection-free algorithms, such as the sto…
University of Washington（华盛顿大学）；Rutgers University（罗格斯大学）；University of Wisconsin-Madison（威斯康星大学麦迪逊分校）
Existing results for low-rank matrix recovery largely focus on quadratic loss, which enjoys favorable properties such as restricted strong convexity/smoothness (RSC/RSM) and well conditioning over all low rank matrices. However, many interesting problems involve more general, non-quadratic losses, which do not satisfy such properties. For these problems, standard nonconvex approaches such as rank-constrained projected gradient descent (a.k.a. iterative hard thresholding) and Burer-Monteiro factorization could have poor empirical performance, and there is no satisfactory theory guaranteeing glo…
Florida Institute for Cyber Security (FICS) Research, University of Florida（佛罗里达大学）；IBM Research（IBM研究院）；University of Wisconsin-Madison（威斯康星大学麦迪逊分校）；Air Force Research Laboratory（空军研究实验室）
Designing deep networks robust to adversarial examples remains an open problem. Likewise, recent zeroth order hard-label attacks on image classification models have shown comparable performance to their first-order, gradient-level alternatives. It was recently shown in the gradient-level setting that regular adversarial examples leave the data manifold, while their on-manifold counterparts are in fact generalization errors. In this paper, we argue that query efficiency in the zeroth-order setting is connected to an adversary's traversal through the data manifold. To explain this behavior, we p…
The Ohio State University（俄亥俄州立大学）；The University of Utah（犹他大学）；Duke University（杜克大学）
SARAH and SPIDER are two recently developed stochastic variance-reduced algorithms, and SPIDER has been shown to achieve a near-optimal first-order oracle complexity in smooth nonconvex optimization. However, SPIDER uses an accuracy-dependent stepsize that slows down the convergence in practice, and cannot handle objective functions that involve nonsmooth regularizers. In this paper, we propose SpiderBoost as an improved scheme, which allows to use a much larger constant-level stepsize while maintaining the same near-optimal oracle complexity, and can be extended with proximal mapping to handl…
Nanyang Technological University（南洋理工大学）；Indiana University Bloomington（印第安纳大学布卢明顿分校）
We study the problem of learning to cluster data points using an oracle which can answer same-cluster queries. Different from previous approaches, we do not assume that the total number of clusters is known at the beginning and do not require that the true clusters are consistent with a predefined objective function such as the K-means. These relaxations are critical from the practical perspective and, meanwhile, make the problem more challenging. We propose two algorithms with provable theoretical guarantees and verify their effectiveness via an extensive set of experiments on both synthetic …
Bosch Center for Artificial Intelligence（博世人工智能中心）；College of Information and Computer Sciences（信息与计算机科学学院）；University of Massachusetts Amherst（马萨诸塞大学阿默斯特分校）；Carnegie Mellon University（卡内基梅隆大学）
We study the problem of generating adversarial examples in a black-box setting, where we only have access to a zeroth order oracle, providing us with loss function evaluations. Although this setting has been investigated in previous work, most past approaches using zeroth order optimization implicitly assume that the gradients of the loss function with respect to the input images are \emph{unstructured}. In this work, we show that in fact substantial correlations exist within these gradients, and we propose to capture these correlations via a Gaussian Markov random field (GMRF). Given the intr…
Duke University（杜克大学）；The University of Utah（犹他大学）；The Ohio State University（俄亥俄州立大学）
Zeroth-order optimization (ZO) algorithms have been recently used to solve black-box or simulation-based learning and control problems, where the gradient of the objective function cannot be easily computed but can be approximated using the objective function values. Many existing ZO algorithms adopt two-point feedback schemes due to their fast convergence rate compared to one-point feedback schemes. However, two-point schemes require two evaluations of the objective function at each iteration, which can be impractical in applications where the data are not all available a priori, e.g., in onl…
Institute for Interdisciplinary Information Sciences, Tsinghua University（清华大学交叉信息研究院）；Department of Mathematics and Joint Center for Quantum Information and Computer Science, University of Maryland（马里兰大学数学系及量子信息与计算机科学联合中心）；Center on Frontiers of Computing Studies, Peking University（北京大学计算前沿研究中心）；Center for Theoretical Physics, Massachusetts Institute of Technology（麻省理工学院理论物理中心）；Department of Computer Science and Joint Center for Quantum Information and Computer Science, University of Maryland（马里兰大学计算机科学系及量子信息与计算机科学联合中心）
We initiate the study of quantum algorithms for escaping from saddle points with provable guarantee. Given a function $f\colon\mathbb{R}^{n}\to\mathbb{R}$, our quantum algorithm outputs an $ε$-approximate second-order stationary point using $\tilde{O}(\log^{2} (n)/ε^{1.75})$ queries to the quantum evaluation oracle (i.e., the zeroth-order oracle). Compared to the classical state-of-the-art algorithm by Jin et al. with $\tilde{O}(\log^{6} (n)/ε^{1.75})$ queries to the gradient oracle (i.e., the first-order oracle), our quantum algorithm is polynomially better in terms of $\log n$ and matches it…
Double Variance Reduction: A Smoothing Trick for Composite Optimization Problems without First-Order Gradient
School of Management, Xi’an Jiaotong University（西安交通大学管理学院）；State Grid Corporation of China（中国国家电网有限公司）；International Business School, Beijing Foreign Studies University（北京外国语大学国际商学院）；Agency for Science, Technology and Research (A*STAR)（新加坡科技研究局）；College of Computing and Data Science, NTU（南洋理工大学计算机与数据科学学院）
Gaussian Process Bandit Optimization with Machine Learning Predictions and Application to Hypothesis Generation
Department of Management Science and Engineering, Stanford University, Stanford, USA（管理科学与工程系，斯坦福大学，斯坦福，美国）；Graduate School of Business, Stanford University, Stanford, USA（商学院，斯坦福大学，斯坦福，美国）
MoSSP: A Momentum-Based Single-Loop Stochastic Penalty Method for Nonconvex Constrained DC-Regularized Optimization
School of Mathematical Sciences, Beihang University, Beijing 100191, China（北京航空航天大学数学学院）；School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou 510006, China（中山大学计算机科学与工程学院）
