Skip to main content
Auto-learning lets the system learn from operators’ decisions during document processing to improve field detection. When the system cannot find a field, an operator indicates its correct location. After the corrected documents are exported, the system uses those corrections as learning input. Field extraction training can also be done by the administrator if a project must be trained before operators start working on it.

​
Configure auto-learning

1

Create the Document Definition

Create a Document Definition. In its section properties, select Allow field location training.
2

Create the fields

Create the fields in the section, and select Can have region in each field’s properties.
3

Save and publish

Save and publish the Document Definition.
4

Enable the Training stage

In the batch type properties, click Workflow and enable the Training stage.
After the administrator sets up the Document Definition, the system automatically learns from the operators’ corrections on the Verification Stations.

​
Configure variants for auto-learning

For documents of the same type whose appearance varies greatly, create variants for each field layout and train a classifier to distinguish them. For more information, see Variable field locations on documents of the same type.
1

Add section variants

Add section variants using one of the following methods:
  • Create variants manually: on the Data Sets tab of the section properties, click View, then Add.
  • Load variants from a database: on the Data Sets tab, click Set Up, and select Database as the data source.
  • Create variants using a script: on the Data Sets tab, click Set Up, and select Script as the data source.
2

Save and publish

Save and publish the Document Definition.
3

Train a classifier on the variants

  1. Switch to Open Classifier Training Batches mode and load document images into a new batch.
  2. Assign a reference class to each document, using variants as separate classes. Click Set Class → Add → Add, select Specify variant, and then select a variant from the list.
  3. Train the classifier with Project → Classification Training → Train.
To let operators check and change the variant assigned to a document, display the variant IDs on the form with a service field. For more information, see Enable operators to change variants.

​
How auto-learning works

Documents whose field locations operators have verified and corrected are placed into a training batch and matched against the current version of the trained FlexiLayout. If all fields are found correctly, the FlexiLayout does not need retraining. Sometimes the FlexiLayout finds the fields correctly even though the operator had to change them: the documents were processed with an old or untrained version while awaiting verification, and the system trained the FlexiLayout on other documents in the meantime. These documents stay in the training batch with For testing status and are used for regression tests to keep future versions from degrading. If a trained FlexiLayout is applied and some field regions do not match, those documents are used to train a new version and are assigned For training status. The result is a new FlexiLayout version. FlexiCapture applies both the new and previous versions to the For training and For testing documents, checks how well the detected field regions match what the user confirmed, keeps the version with the best match, and deletes the other.