20240037392. FURTHER TRAINING OF NEURAL NETWORKS FOR THE EVALUATION OF MEASUREMENT DATA simplified abstract (Robert Bosch GmbH)

From WikiPatents
Jump to navigation Jump to search

FURTHER TRAINING OF NEURAL NETWORKS FOR THE EVALUATION OF MEASUREMENT DATA

Organization Name

Robert Bosch GmbH

Inventor(s)

Frank Schmidt of Leonberg (DE)

FURTHER TRAINING OF NEURAL NETWORKS FOR THE EVALUATION OF MEASUREMENT DATA - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240037392 titled 'FURTHER TRAINING OF NEURAL NETWORKS FOR THE EVALUATION OF MEASUREMENT DATA

Simplified Explanation

The abstract describes a method for further training a pre-trained neural network that processes measurement data. The method involves providing a batch of new training examples and a subset of previous training examples. The neural network processes these examples and evaluates the deviations of the outputs from the target outputs using a predefined cost function. The parameters of the neural network are then optimized to improve the evaluation with the cost function for the new training examples without worsening the evaluation for the previous training examples.

  • The method involves further training of a pre-trained neural network using new and previous training examples.
  • A batch of new training examples and a subset of previous training examples are provided.
  • The neural network processes these examples and evaluates the deviations of the outputs from the target outputs.
  • A predefined cost function is used to evaluate the deviations.
  • The parameters of the neural network are optimized to improve the evaluation for the new training examples and maintain the evaluation for the previous training examples.

Potential Applications:

  • This method can be applied in various fields where neural networks are used for processing measurement data, such as image recognition, natural language processing, and sensor data analysis.
  • It can be used in industries like healthcare, finance, manufacturing, and transportation to improve the accuracy and performance of neural network models.

Problems Solved:

  • The method solves the problem of improving the performance of a pre-trained neural network when processing new training examples.
  • It addresses the challenge of optimizing the behavior of the neural network to minimize deviations from target outputs.
  • It allows for continuous learning and improvement of the neural network model without compromising the performance on previous training examples.

Benefits:

  • The method enables the neural network to adapt and improve its performance with new training examples.
  • It enhances the accuracy and reliability of the neural network model by optimizing its parameters.
  • It provides a cost-effective approach to further train pre-existing neural network models without the need for retraining from scratch.


Original Abstract Submitted

a method for further training of a neural network for processing measurement data, which neural network has been pre-trained with training examples from a set m. in the method: a batch b of new training examples is provided; a subset d⊆m of the previous training examples is provided; the new training examples from batch b and the previous training examples from subset d are processed by the neural network into outputs respectively; the deviations of the outputs from the respective target outputs are evaluated using a predefined cost function; parameters characterizing the behavior of the neural network are optimized with the aim that, during further processing of previous and new training examples, the evaluation with the cost function is improved in regard to new training examples from batch b and is not made worse in regard to previous training examples from subset d.

Retrieved from "http://wikipatents.org/index.php?title=20240037392._FURTHER_TRAINING_OF_NEURAL_NETWORKS_FOR_THE_EVALUATION_OF_MEASUREMENT_DATA_simplified_abstract_(Robert_Bosch_GmbH)&oldid=30791"