Back to results
US 12,645,960 B2Granted
Learning apparatus, method and inference apparatus
AssigneeKabushiki Kaisha Toshiba
InventorsTomohiro Yamasaki, Yoshiyuki Kokojima
Life of this patent
Estimating the term…
Abstract
According to one embodiment, a learning apparatus includes a processor. The processor acquires a document to which a tag is added. The processor converts the document into an image to generate a document image, and converts the tag into an image according to composition of the document image to generate a tag image. The processor trains a network model using the document image as input data and the tag image as ground truth data to generate a trained model.
Claims16
A learning apparatus comprising a processor configured to:
- acquire a document in which each specific range of text data is associated with a tag, the document including the text data;
- convert the document into an image to generate a document image, and convert, based on types of the tags and areas of the tags, each of the tags into an image according to composition of the document image to generate tag images;
- and
- repeatedly train a network model and update parameters of the network model using the document image as input data and the tag images as ground truth data to learn whether or not each pixel of the document image is an area to which a tag is added, to generate a trained model having a more robust inference performance.
1 more element
A learning method comprising:
- acquiring a document in which each specific range of text data is associated with a tag, the document including the text data;
- converting the document into an image to generate a document image;
- converting, based on types of the tags and areas of the tags, each of the tags into an image according to composition of the document image to generate tag images;
- and
- repeatedly training a network model and updating parameters of the network model using the document image as input data and the tag images as ground truth data to learn whether or not each pixel of the document image is an area to which a tag is added, to generate a trained model having a more robust inference performance.
2 more elements
Dependencies are inferred from the claim text.