Deep Progressive Search for Electromagnetic Structure Design Under Limited Evaluation Budgets
Keywords: Electromagnetic Structure, Surrogate Model, Tree Search
TL;DR: We introduce a method to efficiently design electromagnetic structures under limited simulation budgets by employing design space management and a consistency-based sample selection strategy.
Abstract: Electromagnetic structure (EMS) design aims to optimize a material distribution, e.g., metals over a printed circuit board, which is crucial for antenna and meta-material. This task, however, is inherently a highly non-convex problem with no explicit objective function, making it extremely challenging to solve. The most common approach to addressing this problem relies on evolutionary algorithms (e.g., Genetic Algorithm), where candidate structures are evaluated through electromagnetic simulation using specialized software. However, these methods struggle with inefficiency, especially when dealing with large structural design space and time-consuming simulations.
To address this, we propose a Deep Progressive Search method called DPS, which leverages a Deep Neural Network (DNN) as a surrogate model to identify a satisfactory structure within a limited simulation budget. Specifically, we develop a tree-search-based design space control strategy that models the design space as a tree and incrementally refines it through node expansions, enabling adaptive exploration of more complex regions while leveraging insights from simpler subspaces. Moreover, we introduce a consistency-based sample selection strategy to balance exploration and exploitation. Experiments on two real-world engineering tasks, i.e., Dual-layer Frequency Selective Surface and High-gain Antenna show the effectiveness of the proposed DPS in terms of efficiency under limited evaluation budgets.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
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.
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: 2900
Loading