Skip to main content
The Fast Learning activity is used to extract fields from structured and semi-structured documents. It also allows training the fields selected as output fields for this activity while documents are being processed in Vantage. For more information, see Vantage Runtime Guide, Online Learning. You can explicitly disable field training by deselecting it on the Activity Properties pane. If you do not add the Fast Learning activity to your document processing flow, it will not be possible to train fields once you have created and published your skill. If you are editing a skill created in Vantage, the skill may contain a pre-trained Fast Learning activity. You can add other activities and combine them with the pre-trained one. For more information, see Editing a skill created and trained in Vantage.
Note: The Fast Learning activity can’t extract complex structures (e.g. nested tables, which are repeating structures inside other tables) and fields of type Image. To extract such structures, use Extraction Rules activity.

​
Use Cases

Add this activity to your document processing flow in the following cases:
  • When the document set includes several document layout variants, and you are able to provide samples for each variant during training. For example, if you want to train extraction from bank statements from several different banks, and you have samples from each bank at your disposal.
  • When you are planning to process document variants on which your skill has not yet been trained, and you want to benefit from Online Learning. For example, when processing invoices, each supplier likely has their own invoice layout, moreover, new suppliers can appear daily. In this case, you will use other activities to extract data from the documents, but you can also add the Fast Learning activity to the processing flow, and it will be trained during runtime using Online Learning feedback from the manual review loop.
  • When you want to train fields while documents are being processed in Vantage.

​
How It Works

Fast Learning is based on a clustering technology that groups similar-looking document layouts together and internally trains a field extraction model for each cluster. The Fast Learning activity can learn thousands of different document variants. As opposed to the Deep Learning activity, the Fast Learning activity tends to memorize what it has “seen” rather than learn image patterns. Fast Learning will not be able to generalize to new document variants it has not yet encontered. When a Fast Learning activity faces a new document during runtime, it establishes which cluster the document is most similar to, and then applies the corresponding internal model. This activity does not require a large training set—one document is enough to start training. If you have several variants of the same document (for example, documents that are essentially identical but look somewhat different), we recommend including documents representing each different variant in the training set. For more information, see Setting up a Fast Learning activity.