Huawei Develops Advanced Neural Network Model Training Method

Strategic Insights -

Huawei Technologies Co., Ltd. has been assigned a patent (No. US 12406488 B2) developed by inventors Chuanyun Deng, Guanfu Chen, and Shaohua Tang for "Neural network model training method, image processing method, and apparatus." This breakthrough in artificial intelligence involves a novel approach to training neural network models, which has significant implications for the field of deep learning.

Key Takeaways:

  • The patent describes a neural network model training method that involves inputting training data for feature extraction and obtaining a first weight gradient of the neural network model based on the extracted feature.
  • The method also includes obtaining a candidate weight parameter, where the partial derivative of the function value of a target loss function to the candidate weight parameter is 0.
  • The apparatus disclosed in the patent includes a neural network model training system that utilizes the novel training method to achieve improved accuracy and efficiency in image processing.
  • The inventors, Chuanyun Deng, Guanfu Chen, and Shaohua Tang, hail from Hangzhou, China, and their development is expected to have a significant impact on the field of artificial intelligence.
  • The patent application was initially filed on March 9, 2023, and has been assigned the patent number US 12406488 B2.

Statistics:

  • The patent was assigned on September 2, an unspecified year.
  • The patent describes a neural network model training method that achieves improved accuracy and efficiency in image processing.
  • The method involves inputting training data for feature extraction, which is performed using a neural network model.
  • The neural network model training system has the potential to revolutionize the field of artificial intelligence, particularly in the areas of computer vision and natural language processing.

Sources:

  • HUAWEI TECHNOLOGIES CO., LTD., Shenzhen, China (Source of patent abstract)
  • US Patent and Trademark Office (assigned patent number US 12406488 B2)
  • [Original patent document] (date unspecified)