9 Human In The Loop

Xin Zou редактировал(а) эту страницу 2020-12-17 12:26:39 -08:00

Убрать экранирование Экранировать

Содержание

• HUMAN IN THE LOOP (HITL)
• How does HITL work?
• Scenario 1: Analyze a local file and then add it to training set
• Scenario 2: Auto-label current document

Этот файл содержит неоднозначные символы Юникода!

Этот файл содержит неоднозначные символы Юникода, которые могут быть перепутаны с другими в текущей локали. Если это намеренно, можете спокойно проигнорировать это предупреждение. Используйте кнопку Экранировать, чтобы подсветить эти символы.

HUMAN IN THE LOOP (HITL)

Human in the loop(HITL) is a way to speed up the labeling process and improve model quality.

When you have a lot of documents to label, and analyze a lot of testing files to improve model quality, it's a long process. HITL can help you here.

How does HITL work?

Scenario 1: Analyze a local file and then add it to training set

1. Train a model with 5 labeled documents
2. Pick a up local file and analyze it, if you find there are incorrect fields, you could add the document to training set then adjust the labels if necessary.
3. Train a new model with new labeling data.
4. Repeat step 2 - 3 until the model is good enough.

Scenario 2: Auto-label current document

1. Upload lots of documents into your Azure blob storage, train a model with at least 5 labeled documents.
2. Click auto-label, FoTT will automatically pick up other files in the blob storage and analyze them, then return the analyze results as label files. (note: there will be cost associated with using Form Recognizer Analyze API, please be aware).
3. Users can revise the labels of those files if necessary, notice the icons of the document are changed, indicating it’s been revised and ready to be used for training.
4. You could start a new training session with newly revised files.
5. Repeat step 2 - 4 until the model is good enough.