About ML Packages

Using a Document Understanding ML Package involves these steps:

• Collecting document samples and the requirements of the data points that need to be extracted.
• Labeling documents using Document Manager. Document Manager itself connects to an OCR Service.
• Downloading or exporting labeled documents as a Training dataset and uploading the exported folder to AI Center Storage.
• Downloading or exporting labeled documents as an Evaluation dataset and uploading the exported folder to AI Center Storage.
• Running a Training Pipeline on AI Center.
• Evaluating the model performance with an Evaluation Pipeline on AI Center.
• Deploying the trained model as an ML Skill in AI Center.
• Querying the ML Skill from an RPA workflow using the UiPath.DocumentUnderstanding.ML activity package.
  Note: Remember that using Document Understanding ML Packages requires that the machine on which AI Center is installed can access https://du-metering.uipath.com.
  Important: When creating a UiPath.DocumentUnderstanding.ML.Activities Package in AI Center, the package name should not be any python reserved keyword, such as class , break, from, finally, global, None, etc. Note that this list is not exhaustive since the package name is used for class <pkg-name> and import <pkg-name> .

These are out-of-the-box Machine Learning Models to classify and extract any commonly occurring data points from semi-structured or unstructured documents, including regular fields, table columns, and classification fields, in a template-less approach.

Note:

Out-of-the-box Machine Learning Packages that are delivered by UiPath have version 0 and are already available on your tenant, meaning that there is no need to download them.

Download is available only for versions 1 or higher, that were already trained by you.

Document Understanding contains multiple ML Packages split into five main categories:

• UiPathDocumentOCR
• DocumentUnderstanding
• DocumentClassifier
• Out-of-the-box Pre-trained ML Packages
• Out-of-the-box Pre-trained DU ML Packages in AI Center
• Other Out-of-the-box ML Packages

UiPathDocumentOCR

This is a non-retrainable model which can be used with the UiPath Document OCR engine activity as part of the Digitize Document activity. To be used, the ML Skill must first be made public so that a URL can be copy-pasted into the UiPath Document OCR engine activity.

You can run UiPathDocumentOCR on GPU or CPU, accuracy being the same on both cases, training on GPU being faster than te one on CPU. Check the Hardware Requirements for more details.

UiPathDocumentOCR requires access to the Document Understanding metering server at https://du.uipath.com/metering if the ML skill is running on an AI Center on-premises regular deployment. No internet access is needed on AI Center on-premises air-gapped deployments.

UiPathDocumentOCR_CPU

This ML Package can be deployed the same way as the UiPathDocumentOCR ML Package, with the following differences:

• it is optimized to run on CPU, so you should see a 3-4x speedup when running in workflow, and 5-10x speedup when using it to import documents into Document Manager
• accuracy is slightly lower than the UiPathDocumentOCR ML Package, and it is similar to the UiPath.DocumentUnderstanding.OCR.LocalServer Studio package
• due to being faster, the CPU is also recommended when documents are large (over 20 pages per doc) in the absence of a GPU, which is ideal.

DocumentUnderstanding

This is a generic, retrainable model for extracting any commonly occurring data points from any type of structured or semi-structured documents, building a model from scratch. This ML Package must be trained. If deployed without training first, deployment fails with an error stating that the model is not trained.

DocumentClassifier

This is a generic, retrainable model for classifying any type of structured or semi-structured documents, building a model from scratch. This ML Package must be trained. If deployed without training first, deployment fails with an error stating that the model is not trained.

Out-of-the-box Pre-trained ML Packages

These are retrainable ML Packages that hold the knowledge of different Machine Learning Models.

They can be customized to extract additional fields or support additional languages using Pipeline runs. Using state-of-the-art transfer learning capabilities, this model can be retrained on additional labeled documents and tailored to specific use cases or expanded for additional Latin, Cyrillic or Greek language support.

The dataset used may have the same fields, a subset of the fields, or have additional fields. To benefit from the intelligence already contained in the pre-trained model, you need to use fields with the same names as in the out-of-the-box model itself.

These ML Packages are:

• Invoices: The fields extracted out-of-the-box can be found here.
• InvoicesAustralia: The fields extracted out-of-the-box can be found here.
• InvoicesIndia: The fields extracted out-of-the-box can be found here.
• InvoicesJapan Preview: The fields extracted out-of-the-box can be found here.

  Retraining using data from Validation Station is currently not supported.

• InvoicesChina Preview: The fields extracted out-of-the-box can be found here.

  Retraining using data from Validation Station is currently not supported.

• Receipts: The fields extracted out-of-the-box can be found here.
• Purchase Orders: The fields extracted out-of-the-box can be found here.
• Utility Bills: The fields extracted out-of-the-box can be found here.
• ID Cards : The fields extracted out-of-the-box can be found here.
• Passports: The fields extracted out-of-the-box can be found here.
• RemittanceAdvices: The fields extracted out-of-the-box can be found here.
• BillsOfLading: The fields extracted out-of-the-box can be found here.
• W2: The fields extracted out-of-the-box can be found here.
• W9: The fields extracted out-of-the-box can be found here.
• ACORD125: The fields extracted out-of-the-box can be found here
• I9: The fields extracted out-of-the-box can be found here
• 990 Preview: The fields extracted out-of-the-box can be found here
• 4506T: The fields extracted out-of-the-box can be found here
• FM1003 Preview: The fields extracted out-of-the-box can be found here
• ACORD25 - The fields extracted out-of-the-box can be found here
• 1040 - The fields extracted out-of-the-box can be found here
• Checks - The fields extracted out-of-the-box can be found here
• Bank Statements - The fields extracted out-of-the-box can be found here
• Financial statements - The fields extracted out-of-the-box can be found here
• Packing Lists - The fields extracted out-of-the-box can be found here
• ACORD131 - The fields extracted out-of-the-box can be found here
• ACORD126 - The fields extracted out-of-the-box can be found here
• ACORD140 - The fields extracted out-of-the-box can be found here
• Vehicle Titles - The fields extracted out-of-the-box can be found here

These models are deep learning architectures built by UiPath. A GPU can be used both at serving time and training time but is not mandatory. A GPU delivers>10x improvement in speed for Training in particular.

The Out-of-the-box Pre-trained ML Packages can be split into document categories based on the intended use of each model:

Document Category	ML Model
KYC	Passports ID Cards Utility Bills 4506T 1040 Checks Vehicle Titles
Insurance	ACORD125 ACORD131 ACORD126 ACORD140
Lending	FM1003 W2 4506T
HR	Passports ID Cards W9 I9
Shipping	Invoices Bills of Lading (includes Sea Waybills and Air Waybills) Packing Lists
AP	Invoices Utility Bills
AR	Remittance Advices Purchase Orders Bank Statements Financial Statements
Expenses	Receipts

Other Out-of-the-box ML Packages

These are non-retrainable Packages that are required for non-ML components of the Document Understanding suite.

These ML Packages are:

• FormExtractor: Deploy as Public Skill and paste the URL into the Form Extractor activity.
• IntelligentFormExtractor: Deploy as Public Skill and paste the URL into the Intelligent Form Extractor activity. Make sure to first deploy the HandwritingRecognition ML Skill and configure that as OCR for the this package.
• IntelligentKeywordClassifier: Deploy as Public Skill and paste the URL into the Intelligent Keyword Classifier activity.
• HandwritingRecognitionOCR: Deploy as Public Skill and use as OCR when creating the IntelligentFormExtractor package.