Model Training document understanding

[Christodoulos (Christodoulos Christodoulou)]

Greetings community .

I want to train the ML extractor that uses the existing public endpoint for receipts with more receipts after i manually validate the extracted data.
Is this possible and how can i do that?

Any input is welcome. thanks!

[adiijaiin (Aditya Jain)]

Hi @Christodoulos

Greetings!

Yes you can create it,
Receipts ML Model : https://docs.uipath.com/document-understanding/automation-cloud/latest/user-guide/receipts-ml-package

Schema for Receipts Model: https://du.uipath.com/ie/receipts/info/model

Public endpoint: https://du.uipath.com/ie/receipts
And you can use API key from your Cloud instance.

Thanks

Happy Automation!

[Anil_G (Anil Gorthi)]

@Christodoulos

Create a project in ai center and select the ml packages as required and upload receipts to dataset and then perform data labelling as required and use train pipeline under pipelines to train the model and create a skill out of it which can be used in your projects

Cheers

[Nitya1 (Nitya Tomar)]

Hi @Christodoulos

Gather a set of additional receipts that you want to use for training the ML extractor. These receipts should cover a diverse range of scenarios and variations to improve the model’s accuracy.

Manually validate and label the extracted data in the additional receipts. Identify the key fields and their corresponding values, such as date, total amount, vendor name, etc. Labeling the data helps create a training dataset with ground truth annotations.

Combine the manually validated receipts with their corresponding labeled data to create a training dataset. This dataset should include the receipts and the corresponding extracted field values.

Use the training dataset to train the ML extractor. You can use tools like UiPath Document Understanding or other machine learning frameworks to train the model. The exact process will depend on the specific ML extractor implementation you are using.

If necessary, you can perform additional iterations of training and validation to improve the model’s performance. This may involve adjusting the model parameters or using techniques like transfer learning.

Once the model has been trained and validated, deploy the updated model to the existing public endpoint for receipts. This allows the ML extractor to utilize the improved model when processing new receipts.

Thanks!!