Surrogate gradient methods for data-driven foundry energy consumption optimization DOI Creative Commons
Shikun Chen, Tim Kaufmann, R.J. Martin

и другие.

The International Journal of Advanced Manufacturing Technology, Год журнала: 2024, Номер 134(3-4), С. 2005 - 2021

Опубликована: Авг. 13, 2024

Abstract In many industrial applications, data-driven models are more and commonly employed as an alternative to classical analytical descriptions or simulations. particular, such often used predict the outcome of process with respect specific quality characteristics from both observed parameters control variables. A major step in proceeding purely predictive prescriptive analytics, i.e., towards leveraging for optimization, consists of, given parameters, determining variable values that output improves according model. This task naturally leads a constrained optimization problem prediction algorithms. cases, however, best available suffer lack regularity: methods gradient boosting random forests generally non-differentiable might even exhibit discontinuities. The these would therefore require use derivative-free techniques. Here, we discuss alternative, independently trained differentiable machine learning surrogate during procedure. While alternatives less accurate representations actual process, possibility employing derivative-based provides advantages terms computational performance. Using benchmarks well real-world dataset obtained environment, demonstrate can outweigh additional model error, especially real-time applications.

Язык: Английский

Inverse Design of Unitary Transmission Matrices in Silicon Photonic Coupled Waveguide Arrays Using a Neural Adjoint Model DOI Creative Commons
Thomas Radford, Peter R. Wiecha, Alberto Politi

и другие.

ACS Photonics, Год журнала: 2025, Номер unknown

Опубликована: Фев. 12, 2025

The development of low-loss reconfigurable integrated optical devices enables further research into technologies including photonic signal processing, analogue quantum computing, and neural networks. Here, we introduce digital patterning coupled waveguide arrays as a platform capable implementing unitary matrix operations. Determining the required device geometry for specific output is computationally challenging requires robust versatile inverse design protocol. In this work present an approach using high speed network surrogate-based gradient optimization, predicting patterns refractive index perturbations based on switching ultralow loss chalcogenide phase change material, antimony triselinide (Sb2Se3). Results 3 × silicon array are presented, demonstrating control both amplitude each transmission element. Network performance studied optimization tools such data set augmentation supplementation with random noise, resulting in average fidelity 0.94 targets. Our results show that perturbation offer new routes achieving programmable operators, or Hamiltonians simulators, reduced footprint compared to conventional interferometer-mesh technology.

Язык: Английский

Процитировано

0
Synergy between AI and Optical Metasurfaces: A Critical Overview of Recent Advances DOI Creative Commons
Zoran Jakšić

Photonics, Год журнала: 2024, Номер 11(5), С. 442 - 442

Опубликована: Май 9, 2024

The interplay between two paradigms, artificial intelligence (AI) and optical metasurfaces, nowadays appears obvious unavoidable. AI is permeating literally all facets of human activity, from science arts to everyday life. On the other hand, metasurfaces offer diverse sophisticated multifunctionalities, many which appeared impossible only a short time ago. use for optimization general approach that has become ubiquitous. However, here we are witnessing two-way process—AI improving but some also AI. helps design, analyze utilize while ensure creation all-optical chips. This ensures positive feedback where each enhances one: this may well be revolution in making. A vast number publications already cover either first or second direction; modest includes both. an attempt make reader-friendly critical overview emerging synergy. It succinctly reviews research trends, stressing most recent findings. Then, it considers possible future developments challenges. author hopes broad interdisciplinary will useful both dedicated experts scholarly audience.

Язык: Английский

Процитировано

2
Surrogate gradient methods for data-driven foundry energy consumption optimization DOI Creative Commons
Shikun Chen, Tim Kaufmann, R.J. Martin

и другие.

The International Journal of Advanced Manufacturing Technology, Год журнала: 2024, Номер 134(3-4), С. 2005 - 2021

Опубликована: Авг. 13, 2024

Abstract In many industrial applications, data-driven models are more and commonly employed as an alternative to classical analytical descriptions or simulations. particular, such often used predict the outcome of process with respect specific quality characteristics from both observed parameters control variables. A major step in proceeding purely predictive prescriptive analytics, i.e., towards leveraging for optimization, consists of, given parameters, determining variable values that output improves according model. This task naturally leads a constrained optimization problem prediction algorithms. cases, however, best available suffer lack regularity: methods gradient boosting random forests generally non-differentiable might even exhibit discontinuities. The these would therefore require use derivative-free techniques. Here, we discuss alternative, independently trained differentiable machine learning surrogate during procedure. While alternatives less accurate representations actual process, possibility employing derivative-based provides advantages terms computational performance. Using benchmarks well real-world dataset obtained environment, demonstrate can outweigh additional model error, especially real-time applications.

Язык: Английский

Процитировано

1