Active Learning Showcase Tutorial – Invoice Processing

UiPath Technical Tutorial

– Introduction

Active Learning in UiPath Document Understanding is a relatively new way to succeed in automating document extraction. Compared to the previously used methods of using Ai Center with additional retraining, which required a lot of manual work and large datasets, the Modern Experience streamlines the process by helping users focus only on the most important part of the automation development process. This reduces time spent on annotation by up to 80% and makes building models more efficient but also more adaptable to the slight modifications that may occur after the process has been deployed.

In this tutorial, I’ll show how the Modern Experience simplifies model training with a dummy use case: processing invoices. From setting up a project to publishing and using it, this step-by-step guide will help you get started quickly.

– Requirements

Before starting, ensure the following prerequisites are met:

• Automation Cloud account with these products enabled:
  • Document Understanding
  • AI Center
• Sample set of minimum 50 invoices (PDFs with varying structures).
• Studio Web or Desktop, with Public Preview feeds enabled if needed.
• Basic familiarity with UiPath Document Understanding.

Step 1: Create a Modern Project

1. Log into your Automation Cloud account and navigate to the Document Understanding section.
2. Select Create Project, and choose the Modern Preview experience.
3. Drag and drop your dummy invoices into the interface for automatic classification, or manually define document types if required.

📌 Pro Tip: Pre-trained document types (e.g., invoices) are automatically annotated, saving time. Validate or correct these annotations as needed.

Step 2: Build Your Models

1. Classification Model
   • Use the Classification Recommendations to train your model to distinguish between document types (e.g., regular invoices vs. credit notes).
2. Extraction Model
   • Open a document type, click Annotate, and start mapping fields like Invoice Number, Total Amount, or Due Date.
   • Active Learning identifies uncertain areas and suggests fields for validation, minimizing manual effort.

🎯 Highlight: Use the recommendations pane to guide your actions, ensuring efficient and accurate training.

Step 3: Measure and Improve

Access the Measure module to evaluate your model’s performance:

• Analyze Metrics: View detailed reports on classification and extraction accuracy.
• Follow Recommendations: Identify weak areas and perform targeted improvements.

🛠 Optimization Tip: Focus on refining fields with low confidence scores, as they often have the most significant impact on overall performance.

Step 4: Publish and Deploy

1. Navigate to Publish, create a project version, and click Deploy.
2. Consumption Options:
   • Use the cross-platform workflow in Studio Web/Desktop to build automation workflows that use your trained model.
   • Alternatively, integrate via Document Understanding REST APIs for advanced customization.

Step 5: Monitor and Iterate

1. Use the Monitor module to track your automation’s performance.
   • Review the Project Performance Dashboard to gain insights into model effectiveness.
   • Dive into Document Audit for granular feedback at the document level.

🔄 Iterative Learning: Feed newly collected annotations back into the model to continuously improve accuracy and adapt to changing document formats.

Conclusion

UiPath Document Understanding’s Active Learning transforms how we train AI models by streamlining annotation, enhancing accuracy, and drastically reducing setup time. Through this tutorial, we showcased how to create, train, and deploy a model to process dummy invoices, demonstrating the potential for real-world applications.

Ready to explore? Start your journey with Active Learning today and unlock the full potential of UiPath Document Understanding.