Memory-Pruning Algorithm for Bayesian Optimization with Strict Computational Cost Guarantees
Keywords: Bayesian Optimization, Gaussian Processes, Embedded Systems, Random Eviction, Computational Constraints
TL;DR: A memory-pruning algorithm allows Gaussian-Process-based Bayesian Optimization to keep learning while strictly fixing the maximum computational and memory resources used.
Abstract: Bayesian Optimization (BO) is a powerful tool for optimizing noisy and expensive-to-evaluate black-box functions, widely used in fields such as machine learning and various branches of engineering. However, BO faces significant challenges when applied to large datasets or when it requires numerous optimization iterations. The computational and memory demands of updating Gaussian Process (GP) models can result in unmanageable computation times. To address these limitations, we propose a new Bayesian Optimization algorithm with memory pruning (MP-BO), which restricts the maximum training data size by acquiring new queries while concurrently removing data points from the training set. This approach guarantees a maximum algorithmic complexity of $\bigO(m^3)$, where $m \ll n$ is a fixed value and $n$ represent the size of the full training set. The pruning strategy ensures reduced and constant memory usage and computation time, without significantly degrading performance. We evaluate MP-BO on synthetic benchmarks and a real neurostimulation dataset, demonstrating its robustness and efficiency in scenarios where traditional BO would fail under strict computational constraints. Our results suggest that MP-BO is a promising solution for applications that require efficient optimization with limited computing resources.
Primary Area: optimization
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7217
Loading