20240028949. REWARD FEEDBACK FOR LEARNING CONTROL POLICIES USING NATURAL LANGUAGE AND VISION DATA simplified abstract (HITACHI, LTD.)

From WikiPatents
Jump to navigation Jump to search

REWARD FEEDBACK FOR LEARNING CONTROL POLICIES USING NATURAL LANGUAGE AND VISION DATA

Organization Name

HITACHI, LTD.

Inventor(s)

Andrew James Walker of Santa Clara CA (US)

Joydeep Acharya of Milpitas CA (US)

REWARD FEEDBACK FOR LEARNING CONTROL POLICIES USING NATURAL LANGUAGE AND VISION DATA - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240028949 titled 'REWARD FEEDBACK FOR LEARNING CONTROL POLICIES USING NATURAL LANGUAGE AND VISION DATA

Simplified Explanation

The patent application describes systems and methods for rewarding a machine learning algorithm. Here is a simplified explanation of the abstract:

  • The system receives an image and a task description in text format.
  • The image is sliced into multiple sub-images.
  • An embedding model is executed to embed the text of the task description and the sub-images.
  • The embedding model generates a distribution for the sub-images based on their relevance to the task description.
  • The system generates a reward from the distribution for the sub-images.

Potential applications of this technology:

  • Improving machine learning algorithms by providing rewards based on relevance to a given task.
  • Enhancing image recognition and understanding capabilities of machine learning models.
  • Assisting in automated image annotation and classification tasks.

Problems solved by this technology:

  • Addressing the challenge of providing rewards to machine learning algorithms.
  • Improving the accuracy and performance of machine learning models by incorporating task relevance.

Benefits of this technology:

  • Enables better training and fine-tuning of machine learning algorithms.
  • Enhances the ability of algorithms to understand and interpret images.
  • Facilitates automated image analysis and processing tasks.


Original Abstract Submitted

example implementations described herein involve systems and methods for providing a reward to a machine learning algorithm, which can include receiving an image, and a task description defined in text; slicing the image into a plurality of sub-images; executing an embedding model to embed the text of the task description and the sub-images to generate a distribution for the sub-images based on relevance to the task description; and generating the reward from the distribution for the sub-images.