UiPath-SAIv1 UiPath Certified Professional Specialized AI Professional v1.0 Questions and Answers

The Quality of Service (QoS) feature in UiPath Communications Mining is designed to monitor and measure the quality of service within a communications channel. It helps businesses understand how well their communications are being managed by evaluating different performance metrics such as response times and the quality of interactions. QoS ensures that communications are aligned with service level agreements and business expectations.

(Source: UiPath Communications Mining documentation)

In UiPath Communications Mining, high coverage ensures that a larger proportion of communications is classified with meaningful labels, meaning fewer communications are sent for manual review and more processes are captured for automation. This leads to more effective automation, reducing the need for human intervention and ensuring that the automation use case is fully realized.

For more details, refer to:

UiPath Communications Mining Performance Metrics: Coverage and Automation

In UiPath's AI Center, a Pipeline can be scheduled for execution either at a specific future date or with a recurring schedule. This allows for flexibility in automating the retraining or execution of machine learning models based on predefined time intervals or specific dates. For example, you can schedule a pipeline to run once on a given date or set it to run daily, weekly, or monthly, depending on the project's needs.

For more information, refer to:

UiPath AI Center Documentation: Pipeline Scheduling

Automation Pipelines in AI Center: Pipeline Execution and Scheduling

UiPath AI Center offers three primary package types: pre-trained, custom training, and fine-tunable models. These are essential for various use cases, from leveraging existing models to training custom ones based on specific data

 The ML Classifier is a document classifier that uses a machine learning model deployed as an ML Skill in AI Center to perform document classification tasks. The ML Classifier can work by default with Invoices, Purchase Orders, Receipts, and Utility Bills, or with custom document types that are trained using the Data Manager and the Machine Learning Classifier Trainer12.

According to the best practice, the ML Classifier is the only recommended classifier to be used when the custom document types are very similar and file splitting is not necessary. This is because the ML Classifier can handle complex and ambiguous cases where the document types are hard to distinguish by rules or keywords, and can also learn from feedback and improve over time. File splitting is not necessary when the documents are single-page or have a consistent number of pages per document type3.

The other options are not correct because they are scenarios where other classifiers, such as the Keyword Based Classifier or the Intelligent Keyword Classifier, can be used in combination with the ML Classifier or instead of it. These classifiers are based on rules or keywords that can identify the document types based on their content or metadata, and can also perform file splitting if the documents are multi-page or have a variable number of pages per document type3.

References: 1: Machine Learning Classifier - UiPath Activities 2: Machine Learning Classifier Trainer - UiPath Document Understanding 3: Document Classification - UiPath Document Understanding

Document Understanding documentation, when using the "Export Now" tab in the Export Files dialog box within the Document Manager, the available options include:

Download to Excel – This allows downloading the dataset locally in an Excel format.

Export to AI Center – This exports the data directly to the AI Center for use in model training or evaluation.

Export All – This option exports all documents regardless of labeling or filtering status.

These features are designed to facilitate exporting labeled data for further processing in AI Center, or locally for offline analysis

 Based on the classification of documents, there are two common types of data extraction methodologies: rule-based data extraction and model-based data extraction1. Rule-based data extraction targets structured documents, while model-based data extraction is used to process semi-structured and unstructured documents1. However, neither of these methods alone can handle the high variety of layouts that some document types may have. Therefore, a hybrid data extraction approach is recommended, which combines the strengths of both methods and allows for more flexibility and accuracy23. A hybrid data extraction approach can use one or more extractors, such as RegEx Based Extractor, Form Extractor, Intelligent Form Extractor, Machine Learning Extractor, or FlexiCapture Extractor, depending on the document type and the fields of interest3. The Data Extraction Scope activity in UiPath enables the configuration and execution of a hybrid data extraction methodology, by allowing the user to customize which fields are requested from each extractor, what is the minimum confidence threshold for a given data point extracted by each extractor, what is the taxonomy mapping, at field level, between the project taxonomy and the extractor’s internal taxonomy (if any), and how to implement “fall-back” rules for data extraction2.

References: 2: Data Extraction Overview 3: Data Extraction 1: Document Processing with Improved Data Extraction

 Entities are additional elements of structured data which can be extracted from within the verbatims. Entities include data such as monetary quantities, dates, currency codes, organisations, people, email addresses, URLs, as well as many other industry specific categories. Entities represent concepts, themes, and intents that are relevant to the business use case and can be used for filtering, searching, and analyzing the verbatims.

References:

Communications Mining - Entities

Communications Mining - Using Entities in your Application

Communications Mining - Configuring Entities

In UiPath's Taxonomy Manager, field rules are used to optimize extraction results and perform automatic validation of extracted data. These rules can be applied to specific fields within the document understanding process to ensure that extracted data meets certain predefined conditions or constraints, such as formats, patterns, or value ranges. This allows for higher accuracy in the extraction process and reduces errors, as invalid data can be flagged or corrected based on these rules.

Field rules help automate the validation of data by setting criteria that extracted data must meet, thus enhancing the extraction quality and ensuring that only valid and structured data is processed further in the automation workflow.

For more details, refer to:

UiPath Taxonomy Manager Documentation: Field Rules in UiPath

UiPath Document Understanding Framework: Taxonomy Manager Overview

A high-level view of the tabs within an AI Center project is as follows:

Dashboard: This tab provides an overview of the project’s status, such as the number of datasets, pipelines, packages, skills, and logs, as well as the AI Units consumption and quota.

Datasets: This tab enables you to upload, view, and manage the datasets that are used for training and evaluating the ML models within the project. A dataset is a folder of storage containing arbitrary files and sub-folders1.

Data Labeling: This tab enables you to upload raw data, annotate text data in the labeling tool (for classification or entity recognition), and use the labeled data to train ML models. It is also used by the human reviewer to re-label incorrect predictions as part of the feedback process2.

ML Packages: This tab enables you to upload, view, and manage the ML packages and package versions within the project. An ML package is a group of package versions of the same package type, and a package version is a trained model that can be deployed to a skill3.

Pipelines: This tab enables you to create, view, and manage the pipelines and pipeline runs within the project. A pipeline is a description of an ML workflow, including the functions and their order of execution, and a pipeline run is an execution of a pipeline based on code provided by the user4.

ML Skills: This tab enables you to deploy, view, and manage the ML skills within the project. An ML skill is a live deployment of a package version, which can be consumed by an RPA workflow using an ML skill activity in UiPath Studio5.

ML Logs: This tab enables you to view and filter the logs related to the project, such as the events, messages, and errors that occurred during the pipeline runs, skill deployments, and skill executions6.

References:

1: About Datasets 2: About Data Labeling 3: About ML Packages 4: About Pipelines 5: About ML Skills 6: About ML Logs

Data Extraction identifies and extracts specific information from documents to be processed by automations. This is a key part of UiPath's Document Understanding Framework, enabling bots to process structured and unstructured documents effectively.

The Explore phase is the second phase of model training in UiPath Communications Mining, which is a solution that enables the analysis of large volumes of text-based communications using natural language processing and machine learning. The Explore phase builds on the foundations of the taxonomy that was created in the Discover phase by reviewing clusters and searching for different terms and phrases. The objective of the Explore phase is to provide each of the labels or entities that are important for the use case with enough varied and consistent training examples, so that the platform has sufficient training data from which to make accurate predictions across the entire dataset. The Explore phase is the core phase of model training, and requires the most time and effort, but also leads to better model performance and accuracy1.

References: 1: Communications Mining - Explore

According to the UiPath documentation1, the generic Document Understanding ML Package supports data extraction from any type of structured or semi-structured documents, building an ML model from scratch. The supported languages for this package are Latin-based languages, Cyrillic languages, Greek left-to-right, and Japanese (Preview). Additionally, the documentation23 also mentions that the package can support Chinese with the use of an OCR that supports that language. Therefore, the correct answer is D.

References: Document Understanding - Supported Languages, Document Understanding - ML Packages, Document Understanding - ML Packages.

In UiPath Document Understanding Cloud APIs, the Generative Classifier is used to classify documents by leveraging a generative AI model. To classify a document as either an "Invoice" or a "Receipt," a POST request must be sent to the /api/framework/projects/{projectId}/classifiers/generative_classifier/classification endpoint. The body of this request should contain prompts specifying the classification types (in this case, "Invoice" and "Receipt") along with their corresponding descriptions. This allows the model to correctly classify incoming documents based on these predefined prompts.

For further reading, refer to:

UiPath Document Understanding API Documentation: Cloud API for Classification

Generative Classifier APIs: UiPath Classification APIs

The Solution Architect is responsible for devising a strategy to prioritize processes during the Business Case and Technical Validation phase. Their role involves assessing technical feasibility, scalability, and business value to determine process prioritization.

When a pipeline is in the running state in UiPath AI Center, the only action that can be performed is to Kill it. This terminates the process immediately, stopping any further processing. Actions like removing or restarting the pipeline can only be performed after the pipeline has stopped.(Source: UiPath Pipeline documentation​

In UiPath Communications Mining, importing a taxonomy should be done before starting model training. The taxonomy, which includes labels and categories, defines how the data will be classified and structured during the training process. It is essential to have a well-defined taxonomy to ensure accurate predictions and classifications. Importing the taxonomy before training allows the model to learn from it, enhancing its performance. Changes to the taxonomy can be made later, but the initial import is crucial at the start of the training phase to guide the model effectively.

(Source: UiPath Docs on Communications Mining)

The "Wait for Classification Validation Task and Resume" activity in UiPath's Document Understanding Framework is primarily used to halt or suspend the workflow until a specified document classification validation task is completed by a human. This activity is part of the broader workflow to ensure that when automatic classification of documents cannot be confidently achieved, a human-in-the-loop (HITL) approach is followed to validate or correct classifications. Once the validation is performed in UiPath's Action Center by a human, the workflow is resumed, ensuring the proper handling of documents that require review and correction.

This is aligned with the design of the Action Center, which is integrated into UiPath's Document Understanding Framework. When dealing with document classification or extraction confidence issues, manual human validation tasks are often required, which is what this activity manages. It facilitates human oversight, preventing the automation from proceeding with potentially incorrect classifications.

Reference from UiPath documentation:

UiPath Action Center explains how humans are involved in validation tasks to handle cases where classification or extraction needs manual review.

Wait for Task and Resume Activity in UiPath Documentation explains how it waits for a task (such as document validation) to be completed in the Action Center before resuming the workflow.

For more details, you can consult the official UiPath documents:

UiPath Document Understanding Framework

Wait for Classification Validation Task and Resume

This functionality ensures that incorrect data processing due to automation can be caught and rectified by a human, improving accuracy in document handling workflows.

UiPath Studio does not allow connecting to multiple source control plugins at the same time. A single version control system (e.g., GIT, TFS, SVN) can be used per project to manage code versions.

The most likely cause of the error is that the dictionary was not initialized. In UiPath, a dictionary must be initialized before assigning values to its keys. If you attempt to add a key-value pair to a dictionary that has not been initialized, you will encounter a runtime error.

The correct initialization can be done as follows:

PinMapping = New Dictionary(Of String, String)

Explanation of Other Options:

A. PinMapping["John Doe"]: This is the correct syntax for accessing or assigning a value to a dictionary key. There is no issue with this syntax.

B. The "John Doe" key not present: While it is true that the key might not exist, this specific error would occur only during a Get operation, not an assignment.

D. PinMapping<"John Doe">: This is incorrect syntax for working with dictionaries in UiPath.

When you create a training dataset for document classification or data extraction, you need to split your documents into two subsets: one for training the model and one for evaluating the model. The training subset is used to teach the model how to recognize the patterns and features of your document types and fields. The evaluation subset is used to measure the performance and accuracy of the model on unseen data. The evaluation subset should not be used for training, as this would bias the model and overfit it to the data1.

The recommended split of documents for training and evaluation depends on the size and diversity of your data. However, a general guideline is to use a 70/30 or 80/20 ratio, where 70% or 80% of the documents are used for training and 30% or 20% are used for evaluation. This ensures that the model has enough data to learn from and enough data to test on. For example, if you have 15 documents per vendor, you can use 10 documents for training and 5 documents for evaluation. This would give you a 67/33 split, which is close to the 70/30 ratio. You can also use the Data Manager tool to create and manage your training and evaluation datasets2.

References: 1: Document Understanding - Training High Performing Models 2: Data Manager - Creating a Dataset

To configure the Classification Station as an attended activity in UiPath, certain inputs are mandatory for proper functionality. These include:

Taxonomy: The schema that defines the structure of document types and fields.

Document Path: The file path of the document to be classified.

Document Object Model (DOM): Generated from the document using the Digitize Document activity, this is a structured representation of the document.

Document Text: The extracted text of the document, also an output from the Digitize Document activity.

These inputs allow the Classification Station to review and validate the classification results, either manually or based on automatic suggestions from previous processes.

The typical machine learning model deployment process in UiPath AI Center usually takes between 10-15 minutes1. This process involves wrapping the model in UiPath’s serving framework and deploying it within a namespace on AI Fabric’s Kubernetes cluster that is only accessible by your tenant1. Please note that the actual time may vary depending on the complexity of the model and other factors.

AI Center - Managing ML Skills (uipath.com)

According to the UiPath documentation, AI Center provides three types of out-of-the-box model types that can be used for different purposes and scenarios1:

Pre-trained: These are models that are already trained on a large and diverse dataset and can be used as-is for inference or prediction. They do not require any additional data or training from the user. Examples of pre-trained models are Language Detection, Sentiment Analysis, and Question Answering.

Custom training: These are models that allow the user to train them on their own data using the AI Center UI or API. They require the user to upload a dataset, configure the training parameters, and monitor the training progress and results. Examples of custom training models are Image Classification, Text Classification, and Named Entity Recognition.

Fine-tunable: These are models that are pre-trained on a general dataset but can be further trained or fine-tuned on a specific dataset provided by the user. They offer the best of both worlds: the benefit of using a pre-trained model and the flexibility of customizing it to the user’s needs. Examples of fine-tunable models are Object Detection, Text Summarization, and Language Translation.

References:

1: Out-of-the-Box Packages

In this scenario, the rules defined in the Taxonomy Manager are as follows:

IsNotEmpty: Ensures that the field is not empty.

EndsWithUSA: Checks if the extracted address ends with "USA".

ContainsNY: Ensures that the extracted address contains "NY".

The address extracted in this case is: "42 W 80th St. West New York, NJ 1234, USA". While the address ends with "USA" (passing the EndsWithUSA rule), it includes "West New York, NJ", which satisfies the ContainsNY rule even though "NY" is part of "New York". However, the exact behavior of the Contains rule can generate a warning message because "NY" is part of a larger string ("New York"). Since this does not constitute an error but simply a rule conflict, the validation can proceed.

Therefore, the most accurate outcome would be a warning, but validation can still be performed

In UiPath's Taxonomy Manager, the "Possible values" rule should be used when a text field can have one of several predefined values. This ensures that the extracted data is validated against a list of acceptable values, helping to maintain consistency and accuracy during the extraction process. It is particularly useful for fields such as status indicators or categorical fields where only a limited set of options is valid.

(Source: UiPath Taxonomy Manager documentation)

The Data Extraction Scope activity provides a scope for extractor activities, enabling you to configure them according to the document types defined in your taxonomy. The output of the activity is stored in an ExtractionResult variable, containing all automatically extracted data, and can be used as input for the Export Extraction Results activity. This activity also features a Configure Extractors wizard, which lets you specify exactly what fields from the document types defined in the taxonomy you want to extract1.

The extractors that can be used for Data Extraction Scope activity are:

Regex Based Extractor: This extractor enables you to use regular expressions to extract data from text documents. You can define your own expressions or use the predefined ones from the Regex Based Extractor Configuration wizard2.

Form Extractor: This extractor enables you to extract data from semi-structured documents, such as invoices, receipts, or purchase orders, based on the position and relative distance of the fields. You can define the templates for each document type using the Form Extractor Configuration wizard3.

Intelligent Form Extractor: This extractor enables you to extract data from semi-structured documents, such as invoices, receipts, or purchase orders, based on the labels and values of the fields. You can define the fields for each document type using the Intelligent Form Extractor Configuration wizard.

Machine Learning Extractor: This extractor enables you to extract data from any type of document, using a machine learning model that is trained on your data. You can use the predefined models from UiPath or your own custom models hosted on AI Center or other platforms. You can configure the fields and the model for each document type using the Machine Learning Extractor Configuration wizard.

References: 1: Data Extraction Scope 2: Regex Based Extractor 3: Form Extractor : Intelligent Form Extractor : Machine Learning Extractor

To organize the steps in the correct order for migrating a dataset from the Document Manager to a Modern Project, I'll analyze the instructions and then match the steps accordingly. Here’s the logical order based on the descriptions:

Step 1: Navigate to and open the project into which you want to import data. Select an already existing custom document type or create a new one.

Step 2: Select "Upload" and choose the ZIP file exported from the classic project. Wait for the upload to finish.

Step 3: Go to the document type you want to export and select "Open document type."

Step 4: From the "Filter documents" drop-down list, select "Training and validation set." Select "Export."

Step 5: Leave "Current search results" selected and fill in a name for your export job. Select "Download."

This should reflect the correct sequential process of migrating a dataset from Document Manager to a Modern Project.

The Configure Extractors Wizard is a tool that enables you to select and customize the extractors that are used for data extraction from documents. It is accessed via the Data Extraction Scope activity, which is a container for extractor activities. The wizard allows you to map the fields defined in your taxonomy with the fields supported by each extractor, and to set the minimum confidence level and the framework alias for each extractor. The wizard is mandatory for the extractor configuration, as it ensures that the extractors are applied correctly to each document type and field1.

References: Configure Extractors Wizard of Data Extraction Scope

In UiPath:

Variables are scoped to a specific workflow or sequence and are used to store data for operations within that scope.

Arguments, on the other hand, are used to pass data between workflows, allowing information to flow in or out of invoked workflows.

In UiPath Communications Mining, Dashboards are fully customizable pages that allow users to display relevant charts, visualizations, and insights for a specific dataset. These dashboards provide critical insights into the performance of models, labels, and other metrics relevant to the communication mining process. Users can create and adjust these dashboards to reflect specific data points, making them a powerful tool for tracking and improving model performance over time.

For more details, refer to:

UiPath Communications Mining Dashboards: Customizing Dashboards

UiPath AI Center Documentation: Dashboards and Visualization

Labels in UiPath Communications Mining are user-defined categories that can be applied to communications data, such as emails, chats, and calls, to identify the topics, intents, and sentiments within them1. Labels are trained using supervised learning, which means that users need to provide examples of data that belong to each label, and the system will learn from these examples to make predictions for new data2. However, not all labels are equally easy to train, and some may require more examples than others to achieve good performance. Labels that have bias warnings are those that have relatively low average precision, not enough training examples, or were labelled in a biased manner3. Precision is a measure of how accurate the predictions are for a given label, and it is calculated as the ratio of true positives (correct predictions) to the total number of predictions made for that label. A label with 100% precision means that all the predictions made for that label are correct, but it does not necessarily mean that the label is well-trained. It could be that the label has very few predictions, or that the predictions are only made on a subset of data that is similar to the training examples. This could lead to overfitting, which means that the label is too specific to the training data and does not generalize well to new or different data. Therefore, labels with 100% precision may still have bias warnings if they lack training examples, because this indicates that the label is not representative of the underlying data distribution, and may miss important variations or nuances that could affect the predictions. To improve the performance and reduce the bias of these labels, users need to provide more and diverse examples that cover the range of possible scenarios and expressions that the label should capture.

References: 1: Communications Mining Overview 2: [Creating and Training Labels] 3: Understanding and Improving Model Performance : [Precision and Recall] : [Overfitting and Underfitting] : Fixing Labelling Bias With Communications Mining

When creating a data labeling session in UiPath AI Center, the key pieces of information required are:

The data labeling session name: A unique identifier for the session.

The dataset: The data that will be used in the labeling session.

For more details, refer to:

UiPath AI Center Documentation: Data Labeling Sessions

In UiPath's Document Understanding Framework, UiPath Document OCR is recommended as the initial OCR engine for processing documents. This OCR option is optimized for extracting text from a wide variety of document types and provides good accuracy with structured, semi-structured, and unstructured documents. It is designed to work seamlessly within the UiPath Document Understanding process and is highly compatible with other UiPath activities and models, making it a preferred choice for starting document processing in most projects.

For more details, refer to:

UiPath Document Understanding Framework: UiPath OCR Engines

UiPath Document OCR: UiPath OCR Guide

Main-ActionCenter is a workflow that allows you to create and manage Document Understanding actions in Action Center, which is a web application that enables human intervention in automation processes. You can use Main-ActionCenter when you want to test your Document Understanding process locally, or when you want to implement an attended process that requires human validation or classification of documents. Main-ActionCenter is not recommended for unattended processes, as they do not involve human interaction.

References: Action Center - Document Understanding activities, Document Understanding Process 22.10 now in General Availability!, How to Start a UiPath Document Understanding Project

The model rating is a proprietary score that assesses the overall health and performance of a model in UiPath Communications Mining. It considers four main factors: balance, underperforming labels, coverage, and all labels. The model rating is a score from 0 to 100, which equates to a rating of ‘Poor’ (0-49), ‘Average’ (50-69), ‘Good’ (70-89) or ‘Excellent’ (90-100). A model rating of 70-90 or better indicates that the model has sufficient training and performs well in all of the most important areas. A model rating of 70-90 or better also means that the model has a balanced and representative training data, a low number of labels with performance issues or warnings, a high coverage of the dataset by informative labels, and a high average precision of all labels.

References: Communications Mining - Model Rating, Communications Mining - Understanding and improving model performance

Making an ML Skill public in UiPath enables it to be accessed externally from the UiPath ecosystem. This can be beneficial if the ML Skill needs to be utilized in external applications, systems, or services beyond UiPath's automation environment. Public access expands the usability of the skill, allowing integration with other systems while maintaining security through managed endpoints.

Mining, it is recommended that users provide a minimum of 25 pinned examples per label to ensure proper training and accurate predictions by the machine learning models. This number allows the platform to have a sufficient variety of examples to generalize and make reliable predictions for each label in real-world scenarios.

The minimum number of pinned examples per label is crucial because it enhances both precision and recall, helping the model effectively differentiate between labels and improving overall model performance. If fewer examples are provided, the model may struggle with generalization and might not perform well in distinguishing between similar or overlapping categories.

This standard of 25 pinned examples is outlined in several UiPath documentation sections and best practices for training models in Communications Mining​

UiPath Documentation

​

UiPath Documentation

​

UiPath Community Forum

For further details, refer to UiPath’s official Communications Mining User Guide on their documentation portal.

The Form Extractor in UiPath is best suited for documents that have a fixed or non-variable format. This means that documents where the layout remains consistent and predictable, such as standardized forms or invoices, are ideal for this extraction method. The Form Extractor uses predefined templates that map data fields based on their positions, making it highly effective for extracting data from documents with little to no variation in layout. It is not well-suited for documents with significant layout variability, which would require a more flexible extraction approach, such as a machine learning extractor.

(Source: UiPath Document Understanding documentation)

his is confirmed in the UiPath documentation, where the "Train" feature guides users from the dataset creation stage through the necessary steps to optimize the training process. The system recommends next best actions based on progress to ensure efficient and focused training​

The Classify Document Scope activity in UiPath is used to classify documents supported by the Document Understanding framework. It primarily works with file formats like PDF, JPG, PNG, and other image-based formats but does not process raw or non-standard file types like .raw.

Export the package from AI Center. This is the first step where you prepare the package to be moved.

On the ML Packages page, click the Import ML Package button. This step is where you start the process of importing the package you've exported.

On the Upload package field, add the zip file downloaded using the Downloading ML Packages procedure. After starting the import process, you will upload the actual package.

Click Create. This is the final step where you finalize the uploading process of your ML package.

Please proceed with these steps in the UiPath AI Center to upload your exported package correctly.

The OOB labeling template is a predefined template that you can use to label your text data for entity recognition models. The template comes with some preset labels and text components, but you can also add your own labels using the General UI or the Advanced Editor. When you add an entity label, you need to fill in the following information:

Name: the name of the new label. This is how the label will appear in the labeling tool and in the exported data.

Input to be labeled: the text component that you want to label. You can choose from the existing text components in the template, such as Date, From, To, CC, and Text, or you can add your own text components using the Advanced Editor. The text component determines the scope of the text that can be labeled with the entity label.

Attribute name: the name of the attribute that you want to extract from the text. You can use this to create attributes such as customer name, city name, telephone number, and so on. You can add more than one attribute for the same label by clicking on + Add new.

Shortcut: the hotkey that you want to assign to the label. You can use this to label the text faster by using the keyboard. Only single letters or digits are supported.

Color: the color that you want to assign to the label. You can use this to distinguish the label from the others visually.

References: AI Center - Managing Data Labels, Data Labeling for Text - Public Preview

The synchronous API calls for UiPath Document Understanding Cloud are ideal when you need real-time interaction and fast feedback, such as when processing small documents (with up to five pages). This approach is beneficial in use cases requiring low latency, but it is less suitable for large datasets or documents due to limitations in speed and page count

To automatically retrain and deploy a Document Understanding Machine Learning (ML) Model in AI Center with data from the Document Validation Action, the steps should be followed in this order:

Send Human in the Loop data to the used Dataset.

This step involves sending the data that has been validated and corrected by human reviewers to the dataset. This data will be used for training the ML model.

Use the Export feature from Document Manager using Scheduled Export.

After the data is reviewed and validated, it needs to be exported from the Document Manager. Scheduled Export automates this process, ensuring the dataset in AI Center is regularly updated with new data.

Run Training/Full Pipeline in AI Center using Scheduled Pipeline with Auto-Retraining enabled.

With the updated data in the dataset, the next step is to run the training or the full pipeline. The use of Scheduled Pipeline with Auto-Retraining ensures that the ML model is automatically retrained with the latest data.

Enable the Auto Update feature in the ML Skill.

Finally, enabling the Auto Update feature in the ML Skill ensures that the newly trained model is automatically deployed, making the improved model available for document understanding tasks.

Following these steps in the specified order allows for a streamlined process of continuously improving the ML model based on human-validated data, ensuring better accuracy and efficiency in document understanding tasks over time.

 The ML Extractor is a data extraction tool that uses machine learning models provided by UiPath to identify and extract data from documents. The ML Extractor can work with predefined document types, such as invoices, receipts, purchase orders, and utility bills, or with custom document types that are trained using the Data Manager and the Machine Learning Classifier Trainer12.

According to the best practice, the ML Extractor should be chosen based on the document types, language, and data quality of the documents being processed. It is important to select an ML Extractor that is specifically trained or optimized for the document types that are relevant for the use case, as different document types may have different layouts, fields, and formats. It is also important to take into account the language of the documents, as some ML Extractors may support only certain languages or require specific language settings. Moreover, it is important to consider the quality and diversity of the training data used to train the ML Extractor, as this may affect the accuracy and reliability of the extraction results. The training data should be representative of the real-world data, and should cover various scenarios, variations, and exceptions3.

References: 1: Machine Learning Extractor - UiPath Activities 2: Machine Learning Classifier Trainer - UiPath Document Understanding 3: Data Extraction - UiPath Document Understanding

According to the UiPath documentation, the available options for Scoring in Document Manager, that apply only to string content type, are exact match and Levenshtein. Exact match is a scoring strategy that considers a prediction to be correct only if it exactly matches the true value. Levenshtein is a scoring strategy that measures the similarity between two strings by counting the minimum number of edits (insertions, deletions, or substitutions) required to transform one string into another. The lower the Levenshtein distance, the higher the score. These options can be configured in the Advanced tab of the Edit Field window for string fields.

References:

Document Understanding - Create and Configure Fields

Document Understanding - Training High Performing Models