Search arXiv⌕ Search

arXiv · 2611.10296

Sample complexity and scalability of adaptive-sampling trust-region methods for zeroth-order stochastic optimization --- Part I: common random numbers meet supermartingales

Abstract

We analyze adaptive sampling stochastic model-based trust-region methods that build their models in random subspaces and use common random numbers (CRN) within an iteration, and obtain, within a STORM-type analysis, the first $\tilde{\mathcal{O}}(\varepsilon^{-4})$ bound on the number of function evaluations needed to reach $\varepsilon$-stationarity, both almost surely and in expectation. The subspace model reduces the per-iteration computational cost, which allows scalability. The analysis relies on a general framework for the complexity of adaptive stochastic processes that carries the almost-sure analysis of ASTRO-DF into the STORM setting and adds expected bounds by a direct counting argument, without the renewal-reward argument of Blanchet et al. (2019). The bounds are obtained for two variants of a stochastic model-based trust-region method that differ in how the function evaluations used to measure progress are generated. The first algorithm, ASTRA-F, proceeds in a STORM-like manner: the trial point and the incumbent are re-evaluated on an independent set of replications, so that the estimates entering the acceptance test are independent of the model, as every existing STORM-type analysis requires their accuracy to be assessed conditionally on the model. The second, ASTRA-R, allows a practical approach: the trial point is evaluated on the replications already drawn to build the model, which avoids re-evaluating the incumbent, needs one adaptive sample size per iteration instead of two, and reduces the randomness entering the iteration. To our knowledge, no STORM-type analysis has been able to accommodate the latter approach as the accuracy of an estimate at a point chosen from the samples cannot be assessed conditionally on the model; we overcome this with a probabilistic chaining argument.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Joseph Dzahini, Sara Shashaani, Coralia Cartis. 2026-11-06. Sample complexity and scalability of adaptive-sampling trust-region methods for zeroth-order stochastic optimization --- Part I: common random numbers meet supermartingales. https://arxiv.org/abs/2611.10296

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

Integer matrices that are not copositive have certificates of less than quadratic complexity

A real symmetric matrix is called copositive if the corresponding quadratic form is non-negative on the closed first orthant. If the matrix fails to be copositive there exists some non-negative certificate for which the quadratic form is negative. Due to the scaling property, we can find such certificates in every neighborhood of the origin but their properties depend on the matrix of course and are hard to describe. If it is an integer matrix however, we are guaranteed certificates of a complexity that is at most a constant times the binary encoding length of the matrix raised to the power 3/2.

math.OC↗

Kernel Metrics and Learning for Borel MDPs: Identifiability and Adaptive Control

We consider a Markov decision process with standard Borel spaces and an unknown transition kernel under the average cost criterion. We do not impose any parametrization on the set of possible kernels. To facilitate our analysis, we first develop implication relations between several topologies on kernels defined by pointwise, continuous, or uniform weak convergence; we then review robustness properties on the space of kernels, and finally we establish compactness conditions on the space of kernels. Building on this regularity analysis, we then present two data-driven identifiability results; the first one being Bayesian and the second one empirical. Our conditions for the Bayesian setting are significantly more relaxed compared with prior work which considered either finite or parametric models, though we do not obtain a rate of convergence. Our analysis is asymptotic and builds on measurability in terms of the tail $σ$-field of the available information. Identifiability results are then used to design near-optimal adaptive control policies which alternate between periods of exploration, where the controller acts according to a policy which is conducive to the identification of the true kernel, and periods of exploitation where the controller's information on the kernel is utilized. We will establish that such policies are near optimal. In summary, our contribution is with regard to the general standard Borel setup where there is no apriori parametric representation.

math.OC↗

Differentiable Nonlinear Model Predictive Control

The efficient computation of parametric solution sensitivities is a key challenge in the integration of learning-enhanced methods with nonlinear model predictive control, as their availability is crucial for many learning algorithms. This paper discusses the computation of solution sensitivities of general nonlinear programs using the implicit function theorem and smoothed optimality conditions treated in interior-point methods (IPM). We introduce sequential barrier quadratic programming (SBQP), which generalizes SQP with an IPM quadratic programming (QP) solver by allowing the barrier parameter to approach a prescribed nonnegative target value, recovering classical SQP when this target is zero. Previous works are limited to convex or unconstrained formulations, or lack an implementation for efficient sensitivity evaluation. The publication is accompanied by an efficient open-source implementation within the `acados` framework, providing both forward and adjoint sensitivities for general optimal control problems, achieving speedups exceeding 3x over the state-of-the-art solvers `mpc.pytorch` and `cvxpygen`.

math.OC↗