Adversarially Robust Optimization with Gaussian Processes

Abstract

In this paper, we consider the problem of Gaussian process (GP) optimization with an added robustness requirement: The returned point may be perturbed by an adversary, and we require the function value to remain as high as possible even after this perturbation. This problem is motivated by settings in which the underlying functions during optimization and implementation stages are different, or when one is interested in finding an entire region of good inputs rather than only a single point. We show that standard GP optimization algorithms do not exhibit the desired robustness properties, and provide a novel confidence-bound based algorithm StableOpt for this purpose. We rigorously establish the required number of samples for StableOpt to find a near-optimal point, and we complement this guarantee with an algorithm-independent lower bound. We experimentally demonstrate several potential applications of interest using real-world data sets, and we show that StableOpt consistently succeeds in finding a stable maximizer where several baseline methods fail.

arXiv commentCorrected typos

Cited by in corpus (29)

• Golem: An algorithm for robust experiment and process optimization
• Federated Bayesian Optimization via Thompson Sampling
• Distributionally Robust Bayesian Optimization
• Min-Max Optimization without Gradients: Convergence and Applications to Adversarial ML
• Noisy-Input Entropy Search for Efficient Robust Bayesian Optimization
• Zeroth-Order Algorithms for Nonconvex Minimax Problems with Improved Complexities
• Corruption-Tolerant Gaussian Process Bandit Optimization
• Adversarial Robustness Guarantees for Classification with Gaussian Processes
• On Lower Bounds for Standard and Robust Gaussian Process Bandit Optimization
• Saddle Point Optimization with Approximate Minimization Oracle
• Mean-Variance Analysis in Bayesian Optimization under Uncertainty
• Stochastic Linear Bandits Robust to Adversarial Attacks
• A Domain-Shrinking based Bayesian Optimization Algorithm with Order-Optimal Regret Performance
• Gradient Free Minimax Optimization: Variance Reduction and Faster Convergence
• Adversarial Robustness Guarantees for Gaussian Processes
• Value-at-Risk Optimization with Gaussian Processes
• Combining Pessimism with Optimism for Robust and Efficient Model-Based Deep Reinforcement Learning
• Efficient Batch Black-box Optimization with Deterministic Regret Bounds
• Bayesian Quadrature Optimization for Probability Threshold Robustness Measure
• Adversarial Attacks on Gaussian Process Bandits
• Active learning for distributionally robust level-set estimation
• Mixed Strategies for Robust Optimization of Unknown Objectives
• Cost-Efficient Online Hyperparameter Optimization
• Direct-Search for a Class of Stochastic Min-Max Problems
• Robust Policy Search for Robot Navigation
• No-Regret Learning in Unknown Games with Correlated Payoffs
• Robust Bandit Learning with Imperfect Context
• Bias-Robust Bayesian Optimization via Dueling Bandits
• Bayesian Optimisation with Formal Guarantees