Space Mapping Optimization using Neural Networks for Efficient Parameter Estimation

Dhruvil Kamleshkumar Kotecha · 2025

Read Space Mapping Optimization using Neural Networks for Efficient Parameter Estimation by Dhruvil Kamleshkumar Kotecha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06899.txt.

Introduction

Estimating the best inputs for an expensive simulation, using a fast neural stand-in, is the project reported by Dhruvil Kamleshkumar Kotecha in a solo 2025 paper. A fine model gives accurate answers but costs heavy computation, while a coarse model runs quickly at lower accuracy. Space mapping is a classic strategy for bridging the two.

Kotecha applies neural network-based space mapping optimization: the coarse model is a neural network. The mapping aligns its inputs with the fine model's, so the cheap model's optimum estimates the expensive one's. The project also reviews earlier neuro-modeling space mapping techniques and uses them to strengthen the process. Tests on the chosen derivative-based test problem aim to show the approach optimizing complex functions efficiently. The broader promise is parameter estimation for costly simulators without paying full price for every trial.

Catalog record

CREATORDhruvil Kamleshkumar Kotecha
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-06899

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Space Mapping Optimization using Neural Networks for Efficient Parameter Estimation Dhruvil Kamleshkumar Kotecha Abstract This project focuses on optimizing input parameters of a partial derivative function of a fine model using Neural network-based Space Mapping Optimization (SMO). The fine model is known for its high accuracy but is computationally expensive. On the other hand, the coarse model is represented by a neural network, which is much faster but less accurate. The SMO approach is applied to bridge the gap between these two models and estimate the optimal input parameters for the fine model. Additionally, this project involves a comprehensive review of previously available Neuro Modeling Space Mapping techniques, which are also used in this project to enhance the optimization process. By utilizing SMO with a neural network-based coarse model, we aim to demonstrate the effectiveness of this method in optimizing complex functions efficiently. The proposed approach of using Neural Network based Space Mapping offers a promising solution to this optimization problem.. I Introduction Optimizing complex models has become increasingly crucial in various fields, such as aerospace, automotive, and civil engineering. Optimization techniques are utilized to enhance the design of products and processes, leading to more efficient and cost-effective solutions. Among these techniques, Space Mapping Optimization (SMO) has shown great potential in optimizing the input parameters of a partial derivative function of a fine model. However, optimizing fine models can be computationally expensive, presenting a significant challenge in the optimization process. In such cases, a coarse model represented by a neural network is often used, which is faster but less accurate. This project aims to explore the use of neural network-based Space Mapping Optimization (SMO) to enhance the optimization of a partial derivative function of a fine model. The objective is to efficiently estimate the optimal input parameters for the fine model by bridging the gap between a fine model and a neural network-based coarse model. The use of neural networks to construct surrogate models for complex physical systems is an efficient model, which can replace computationally expensive fine models while still providing accurate approximations of the system’s behavior. Neurospace modeling can be incorporated into the Space Mapping Optimization (SMO) technique to enhance the optimization process’s efficiency. By utilizing neural networks to represent the coarse model, the computational cost of the optimization process can be reduced while maintaining high accuracy. Furthermore, the use of Neurospace modeling can help overcome challenges associated with traditional surrogate modeling approaches, such as the curse of dimensionality and the need for extensive training data. The project’s primary objective is to demonstrate the effectiveness of using SMO with a neural network-based coarse model to optimize complex functions efficiently. II Background II-A Space Mapping Optimization Space mapping optimization is a well-established optimization technique that is utilized to optimize computationally expensive and complex simulations. Initially developed for optimizing microwave circuits in microwave engineering, this technique has been effectively applied to diverse fields such as electromagnetics, fluid mechanics, and structural optimization. Numerous researchers have contributed to this area, including Bandler et al. (2001) [ 1 ] who introduced the space mapping technique. The fundamental concept of space mapping optimization involves creating a coarse model that is computationally inexpensive and easy to evaluate. The design parameters of the fine model are then mapped onto the coarse model, and the coarse model is optimized utilizing standard optimization methods. The optimization outcomes are then mapped back to the fine model, and the process is iterated until convergence is achieved. One of the primary benefits of space mapping optimization is its ability to significantly reduce the computational time and cost necessary for design optimization. This is especially useful in cases where the fine model requires hours or days to simulate, making it impractical for optimization. By utilizing a coarse model, the optimization process can be accelerated significantly, particularly when dealing with large-scale optimization problems, such as designing complex systems or structures. The design parameters of the coarse and fine models are represented by vectors x_{c} and x_{f} , respectively, and their corresponding model responses are represented by R_{c}(x_{c}) and R_{f}(x_{f}) . Notably, R_{c} can be computed more rapidly than R_{f} , but R_{f} produces more precise results. The primary objective of space mapping optimization is to establish a suitable mapping function, represented by P, that maps the fine model parameter space, x_{f} , to the coarse model parameter space, x_{c} . The mapping function can be expressed as follows [ 2 ] : x_{c}=P(x_{f}) (1) The objective is to find a valid mapping function P that satisfies the following condition: R_{c}(P(x_{f}))\approx R_{f}(x_{f}). (2) Once a valid mapping function P has been determined for the region of interest, the coarse model can be used to carry out rapid and precise simulations within the same region. P acts as a link between the coarse and fine models, allowing the optimization process to proceed more effectively. The optimization process starts by creating the coarse model and training it with appropriate data. The fine model is then defined, and P is initialized. The optimization is performed iteratively until convergence is reached. During each iteration, P is utilized to optimize the coarse model, and the optimization outcomes are mapped back to the fine model. This process is continued until the optimized response of the coarse model is sufficiently close to that of the fine model. Once P has been identified, it can be used to optimize the design parameters of the fine model in a more efficient manner. II-B Use of ANN Artificial neural networks (ANNs) are a powerful technique for approximating measurable functions to a desired level of accuracy. When a deterministic relationship exists between input and output variables, multi-layer feed-forward neural networks can achieve this with superior computational efficiency and accuracy, especially in modeling microwave circuit yield and statistical design. Studies by Zaabab et al. (1995) [ 3 ] and Burrascano et al. (1998) [ 4 ] have demonstrated the benefits of using ANNs to optimize the yield of microwave circuits. ANNs are capable of modeling complex relationships and nonlinearities that other approaches may not capture accurately. Training an ANN involves modifying the internal parameters of the neural network to create an ANN model that best fits the training data. This process involves employing optimization techniques such as backpropagation to update the weights in the network. By identifying the optimal combination of weights and biases, the ANN can accurately map inputs to outputs, making it a powerful tool for prediction and modeling within a designated region of interest. It is important to carefully consider the complexity of the ANN, as overly simple networks may result in underfitting, and overly complex networks may result in overfitting. Choosing an appropriate level of complexity is crucial to achieving accurate results. The ultimate objective of using ANNs is to determine optimal values for the internal parameters of the model, such that the response of the coarse model closely approximates the response of the fine model at all learning points [ 5 ] . The response of the fine model be a_{i}=R_{f}(x_{f_{i}}) and the response from the coarse model be b_{i}=R_{c}(P(x_{f_{i}})) . The coarse model( Neural Network in this case) is trained using the same inputs as the fine model. Optimization is

Read the complete text