How to retrain cutstom ML Model

Help AI Center
1

Hello All,

We have built an automation for invoice processing using Document Understanding (Custom ML Model). The automation worked well initially, but as the layout and fields of invoices have changed, we now need to retrain the ML model.

I have watched YouTube videos and UiPath Academy courses, but I couldn’t fully understand the retraining process. Could someone kindly guide me on how to properly retrain the model with recently changed sample invoices?

Thanks,
:sun:

2

Hi @ADITYA_T,

You can retrain your Custom ML Model in Document Understanding by following these steps:

  • Collect new sample invoices with the updated layout and fields.
  • Upload them to AI Center → Datasets (either update existing dataset or create a new version).
  • Perform Labeling on the new documents (make sure fields are correctly mapped).
  • Once labeling is complete, go to Pipelines and create a new Training pipeline using the updated dataset.
  • After training, evaluate the model and check accuracy.
  • If results are good, deploy the new model version and update it in your DU workflow.

Thanks

3

@ADITYA_T

  1. Upload the documents to datasets and label
  2. Then in training select the dataset containing both old and new docs and the base version minor would be 0 again and retrain the mode

But before all of that if you are not on on prem orchestrator and/or your orchestrator is latest you can use latest ixp for two reasons

  1. AI center is being deprecated and moving forward it would not be ideal way to use
  2. IXP provides an unstructured document mode which is based on LLM where prompting for fields would get data and changes in layout wuld not effect the extraction

Cheers