Open Forum with UiPath MVPs session #3


Hey #UiPathCommunity!

As you might have already seen HERE
:loudspeaker: It’s time for another UiPath MVPs Session!

:rocket: Join the chat on 2020-07-28T15:00:00Z
With whom? @mateusznycz @alp.uguray , moderated by @Ioana_Gligan and @Pablito :raised_hands:
Open Forum Session 2

Our MVPs will talk about:
Document Understanding and AI Fabric

Please join us by registering HERE

Additionally we would like to ask you kindly for the questions related to topic. These would be answered during the session. You may ask them under this post.

:red_circle: Recording from the Web Session Available HERE

:page_facing_up: Answers for the unaswered questions from session HERE


Hello everyone!
First of all I would like to thank you all for joining this event. Hopefully it was interesting to you and provide some educational elements as well :slight_smile:

Again big thanks to @alp.uguray and @mateusznycz for awesome presentation and sharing their knowledge.

As I promised here is the link for recording from the event and the answers for the rest questions which were left to answer:

:movie_camera: Open Forum with UiPath MVPs session #3 - recording :red_circle:

:spiral_notepad: Q&A

I would like to understand how should be approach learning AI that can be fabricated with UiPath?

Please check out the Academy course on AI Fabric.

Are there updates in Document Understanding updating validation station and the Attended automation approach?

Yes, Classification Station - a new activity for validating classification of a file - and in conjunction with the Intelligent Keyword Classifier, new capabilities for “file splitting”: identifying multiple documents within a single file.

What are the most recent updates in Document Understanding that would support handwriting extraction?

Check out the Intelligent Form Extractor - and more updates on this end are coming. But currently only for fixed forms or fixed form parts of documents.

Will UiPath improve ML on Invoices recognition with custom training feature or will they rely more on Abby flexicapture in future?

UiPath ML model is completely independent from Abbyy FlexiCapture - both extraction methods are available to be used when one or the other fit. The custom training feature is already available in our Cloud platform and will soon be available on-prem as well.

How can we use AI fabric on Comunity Edition ?

AI Fabric is only available to Enterprise Trial or Enterprise customers at the moment.

I’ve seen some document understanding videos, but now I’m not sure if it mainly deals with standardized forms pulling data from them even if it is hand written. I was under the impression it could classify documents and extract the information from different versions of the same data (e.g. like invoices or if you have data for a process that comes in many different formats of excel file for example from each client, in each case the data needs to be extracted slightly differently)

Please have go at the Academy course, it describes how to handle document processing use cases pretty well.

If you are reading a signature, what field type to you save it to (If you want the signature as it is)?

A boolean let’s say, and then you can see what coordinates the value has, and use that to crop the image.

I have a question on Document Understanding and Intelligent OCR, does it extract data which is inside nested tables?

There is no support for nested tables yet or group fields yet.

We had hard time extracting table data from pdf documents last year. Does new features help extract pdf table to extract as DataTable?

Yes, please have a look at all the extractors with table extraction capabilities: regex, form, intelligent form, ml…

How to use document understanding in AI fabric?

Please go through the Academy courses of AI Fabric and Document Understanding.

Is there any functionality to “tag” or correct the capture if the required field is captured incorrectly within the UI as a human in the loop agent? Such as mistaking the sales tax with the total amount and correcting it?

Yes, in the validation station. Additionally shortly you will be able to collect the vs data through the Train Extractors Scope, to use the human validated data for retraining your models.

Does document understanding extraction (invoices, receipts) work reliably only in English language?

No, it is trained on multiple languages. It works reliably on European languages, including Slavic languages and Greek as long as the ocr engine you use supports it. So it also depends on the ocr the customer chooses. It does not work on right to left languages and on Asian languages (CJK) even if the ocr works on those.

Need help in extracting pdf table data from financial statements in PDF. Does new Document understanding & AI Fabric help?

Depends on your use case. Please have a look at the available extractors and test them against your use case.

Do we have auto training available in verification station now?

The Validation Station is only a UI tool - to perform training, especially with the upcoming releases, you will need to use the train extractors scope after the validation station.