Document understanding classic project

Help Studio
, , ,
1

I am using the Studio DU template for a classic project. It will require retraining. Can I use the retrain classifiers/retrain extractors instead of training in AI center? can either of these be used within the training process or do i need to use AI Center?

2

Hi @ashton.lalchan ,

In UiPath’s Document Understanding (DU) framework, the retraining of classifiers and extractors typically involves using AI Center, which provides the necessary infrastructure and tools for model training, deployment, and management. However, let’s explore the options and capabilities you have:

Retrain Classifiers and Extractors:

  1. Retrain Classifiers:
  • Usage: This activity allows you to retrain a classifier model using new or updated data. It’s used to improve the accuracy of document classification over time by incorporating feedback and additional examples.
  • Within AI Center: Retraining classifiers generally requires AI Center because it leverages AI Center’s machine learning capabilities to process, train, and deploy models.
  1. Retrain Extractors:
  • Usage: This activity is used to retrain extraction models, like the machine learning extractors, to better identify and extract relevant information from documents.
  • Within AI Center: Similar to classifiers, retraining extractors typically requires AI Center to manage the training data, retrain the models, and deploy the updated models.

Using AI Center:

  • AI Center: AI Center provides a comprehensive environment for managing the lifecycle of your AI models, including training, retraining, deployment, and monitoring.
  • Advantages: Using AI Center ensures that your models are trained with sufficient computational resources, proper data handling, and integration with other AI Center features like versioning and scalability.

Retraining Process Without AI Center:

If you aim to perform retraining without AI Center, you would face limitations as the retraining activities are tightly integrated with AI Center’s capabilities. Currently, UiPath does not support standalone retraining of DU models directly within Studio without using AI Center.

Recommended Approach:

To effectively retrain classifiers and extractors for your DU project, follow these steps:

  1. Collect Training Data:
  • Ensure you have sufficient and representative training data, including new documents and feedback from previous runs.
  1. Upload Training Data:
  • Use AI Center to upload and manage your training data.
  1. Train Models in AI Center:
  • Utilize AI Center’s training pipelines to retrain your classifiers and extractors. You can configure these pipelines to periodically retrain models based on new data.
  1. Deploy Updated Models:
  • Once retraining is complete, deploy the updated models from AI Center to be used in your DU processes in Studio.
3

Hi

You can use train classifier and train extractors activity from Studio for training /labelling data with help of present validation station.

Besure that you will need to schedule the finetune data in Document Manager and run the pipeline in AI center for retraining the ML skill with new data set.

4

@ashton.lalchan

Yes you can use train classifier /extractor in studio to train on new documents…but the trained data is to be uploaded to ai center again and the pipeline needs to run which can be scheduled as well to run at specific intervals also ml skill can be scheduled to upgrade whenever train pipeline runs…

Basically you would be creating the required artifacts for ai center in local and then upload them to ai center and then upgrade the package to include the new training as well in the next models

So there is some part that needs to be done in ai center as well but that can be automatic as well via schedules

Cheers

5

hi @Anil_G thanks for this. do you mind breaking down the action steps to perform this? How would I configure the output to AI center from the Classic project?