Leveraging existing ML Skills for prelabelling

[Nishant_Banka1 (Nishant Banka)]

Hello Community,
From our experience with DU projects, we have seen that DU projects tend to be incremental in nature, with new documents/new variations of documents getting added to the mix for Extraction. AI Center provides an amazing feature of using pre-trained ML Models in the prelabelling settting (adding a screenshot containing Prelabelling URL for pre-trained ML Model for Invoice Extraction)

We feel that users can benefit if we can use existing ML Skills in the prelabelling setting, by using the public URL of the ML Skill in the Prelabelling URL field.

Some scenarios where this could be beneficial -

• When we are labeling a type of document which currently doesn’t have any pre-trained ML Model
• When we have custom fields which are not covered in the pre-trained ML Model schema and have to be labelled individually for each document

I understand that there are considerations like how accurate will the ML Skill work on new variation of documents, but if the ML Skill is able to predict 20%-30% of the fields, then it is 20%-30% less work for the user.

Let me and @usharma know your thoughts on this

Thanks,
Nishant

[sharon.palawandram]

you can already use URLs of ML skills which have been made public in AI Center on-premises or in AI Center Cloud.

This methodology ensures that there are only a few extra fields that require manual labeling.

Configure Prelabelling

[Nishant_Banka1 (Nishant Banka)]

Hi @sharon.palawandram,
Exactly what I was looking for. This is amazing!
Something to try out over the weekend.

Regards,
Nishant