{"id":11138,"date":"2022-12-05T08:59:44","date_gmt":"2022-12-05T13:59:44","guid":{"rendered":"https:\/\/blogs.mathworks.com\/deep-learning\/?p=11138"},"modified":"2024-08-21T08:28:40","modified_gmt":"2024-08-21T12:28:40","slug":"using-ai-for-reduced-order-modeling","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/deep-learning\/2022\/12\/05\/using-ai-for-reduced-order-modeling\/","title":{"rendered":"Using AI for Reduced-Order Modeling"},"content":{"rendered":"The following post is from <\/em>Lucas Garcia<\/em><\/a>, Deep Learning Product Manager at MathWorks.<\/em>\r\n
<\/h6>\r\nThis blog discusses the MathWorks\u2019 presence at NeurIPS 2022<\/a> and my talk on \u2018Using AI for Reduced-Order Modeling\u2019 at the conference.\r\n
<\/h6>\r\n\"MathWorks\r\n
<\/h6>\r\nFigure:<\/strong> The MathWorks team at our booth at NeurIPS 2022\r\n
<\/h6>\r\nFounded in 1987, the Conference on Neural Information Processing Systems (abbreviated as NeurIPS) is one of the most prestigious and competitive international conferences in machine learning. Last week, the MathWorks team was at NeurIPS 2022 in New Orleans for the in-person portion of the conference.\r\n
<\/h6>\r\nDuring the Expo Day at NeurIPS, I presented a talk about \u2018Using AI for Reduced Order Modeling\u2019. This blog post provides an overview of this presentation. If you are interested in learning more, check out the Slides: Using AI for Reduced-Order Modeling<\/a>.\r\n
<\/h6>\r\n\"Presentation\r\n
<\/h6>\r\nFigure:<\/strong> Presentation on Using AI for Reduced-Order Modeling at NeurIPS 2022\r\n
<\/h6>\r\nAdditionally, we had many interesting interactions at the booth on deep learning, reinforcement learning, and interoperability between MATLAB and Python\u00ae. My colleagues Drew<\/a> and Naren<\/a> developed an excellent reinforcement learning demo that showcases learning directly from hardware by using a Quanser QUBE\u2122-Servo 2<\/a> and Reinforcement Learning Toolbox<\/a>.\r\n\r\n
\nhttps:\/\/blogs.mathworks.com\/deep-learning\/files\/2022\/12\/RL_CartPole_Quanser_noaudio.mp4<\/a><\/video><\/div>\r\n
<\/h6>\r\nVideo:<\/strong> Reinforcement learning demo showcasing learning directly from hardware\r\n
<\/h6>\r\n \r\n

What is ROM and why use it<\/strong><\/p>\r\nIf you are an engineer or have ever worked in solving an engineering problem, you have probably tried to explain the behavior of a system using first principles. In such situations, you must understand the system\u2019s physics to derive a mathematical representation. The real value of a first-principles model is that results typically have a clear, explainable physical meaning. In addition, behaviors can often be parameterized.\r\n

<\/h6>\r\nHowever, high-fidelity non-linear models can take hours or even days to simulate. In fact, system analysis and design might require thousands or hundreds of thousands of model simulations to obtain meaningful results. This causes a significant computational challenge for many engineering teams. Moreover, linearizing complex models can result in high-fidelity models that do not contribute to the dynamics of interest in your application. In these situations, AI-based reduced-order models can significantly speed up simulations and analysis of higher-order large-scale systems.\r\n
<\/h6>\r\nReduced Order Modeling<\/a> (ROM) is a technique for reducing the computational complexity or storage requirement of a computer model, while preserving the expected fidelity within a controlled error.\r\n
<\/h6>\r\nEngineers and scientists use ROM techniques to:\r\n
<\/h6>\r\n