Published in Nature magazine! Artificial intelligence based on Shengsi MindSpot
Recently, Huawei, in collaboration with Professor Sun Hao of the Hillhouse Institute of Artificial Intelligence of Renmin University of China, has proposed a physics encoded recursive Convolutional neural network (PeRCNN) based on the MindSporeAI framework. This achievement has been published in NatureMachineIntelligence, a sub journal of Nature, and the relevant code has been opened in the MindSporeFlow code repository of Gitee, an open source community
Recently, Huawei, in collaboration with Professor Sun Hao of the Hillhouse Institute of Artificial Intelligence of Renmin University of China, has proposed a physics encoded recursive Convolutional neural network (PeRCNN) based on the MindSporeAI framework. This achievement has been published in NatureMachineIntelligence, a sub journal of Nature, and the relevant code has been opened in the MindSporeFlow code repository of Gitee, an open source community.
Compared to methods such as physical information neural networks, ConvLSTM, and PDE-NET, PeRCN significantly improves model generalization and noise resistance, and improves long-term inference accuracy by more than 10 times. It has broad application prospects in fields such as aerospace, shipbuilding, and meteorological forecasting.
PDE equation plays a central role in the modeling of Physical system, but in the fields of epidemiology, meteorological science, Fluid mechanics and biology, many of the underlying PDEs have not yet been fully explored.
At present, the existing data-driven models rely on Big data, which is difficult to meet in most scientific problems, and there are also interpretative problems. Physical constrained neural networks (PINNs) use prior knowledge to constrain the training of models and reduce the dependence on data, but the soft constraints of PINNs based on Loss function limit the accuracy of the final results. How to obtain results with high accuracy, robustness, interpretability, and generalization in the absence of valid data is still the direction of academic efforts.
PerCNN Model Architecture
Therefore, Huawei, in collaboration with Professor Sun Hao, has developed a physical coded recursive Convolutional neural network based on the MindSporeAI framework by using the powerful computing power of Shengteng AI, and has realized the accurate approximation of nonlinear PDE.
The PeRCNN neural network enforces the encoding of physical structures, and through symbol computation, the underlying fundamental physics expressions can be further extracted from the learned model. This enables PeRCNN to serve as an effective tool to help people accurately and reliably discover potential physical laws from imperfect and high noise data.
In Fluid mechanics, meteorology, oceanography and other disciplines, there are strong nonlinear phenomena such as turbulence and shock waves. The solution of traditional numerical methods requires a lot of computing resources. At present, AI has shown great potential in aircraft flow field, medium-term weather forecast and other issues. PeRCNN has the characteristics of high precision, strong generalization and strong noise resistance. It is expected to break through traditional computing bottlenecks in these fields and accelerate industrial simulation and design, Becoming a new weapon in the field of AI+scientific computing.
Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])