Can we build and utilize deep learning models from Qlik Sense?

[taka1]

Hi,
From my understanding from reading the manual, Qlik AutoML currently supports binary classification, multi classification, and regression model algorithms. Deep learning does not seem to be supported.

Is there any use case for Qlik Sense like the one below? Or is it feasible?
1) Users create their own code to build deep learning models.
2) Users prepare training data in Qlik Sense.
3) The user passes the training data from Qlik Sense to the above code to start building the deep learning model.
4) Utilize this model from Qlik Sense.

Thank you

[igoralcantara]

I see ANN and DL models more suitable for problems like image processing, advanced OCR, language problems not applicable to LLMs and similar problems.

I know some people use DL for classification or regression but I believe this is a terrible idea.

What do you have in mind that would require AutoML to support ANN?

[taka1]

Thank you for your comment.

I would like to make a few corrections to my question. What I would like to know is as follows:

Is it possible to use Qlik Sense to fine tune(*) LLMs?
Also, is it possible to utilize that LLM from Qlik Sense?

The specific steps are as follows;
1) Create your own code to fine-tune the LLM.
2) Prepare training data in Qlik Sense.
3) Pass the training data from Qlik Sense to the above code to fine-tune the LLM.
4) User uses this model from Qlik Sense.

(*) Fine-tuning is the process of taking a pre-trained model and further training it on a domain-specific dataset; in order to take advantage of the natural language processing capacity of pre-trained and open-source LLMs and further train them to perform your specific tasks.

A potential prospect retunes local LLMs with his own code and provides applications that take advantage of this. He wants to know if Qlik Sense can streamline this development effort.

Thank you

[igoralcantara · Accepted solution]

Hi @taka1,

I believe there is a bit of confusion of what is Qlik AutoML. Qlik AutoML performs classic Machine Learning work, meaning, we get structured data and we train models that predicts the values of a given variable. The prediction can be of 3 types so far: Classification (Binary and Multiclass) and Regression.

Fine tune a LLM model is not a classic Machine Learning problem but a language problem. I know some authors consider Transformers a type of Machine Learning but in really they are a different kind of AI.

Having said that, what you need is a RAG system. Qlik has another tool for that, it is called Qlik Answers. I even wrote a few articles on Data Voyagers (datavoyagers.net) about it. With Qlik Answers, you can build a knowledge base with your unstructured data, index it and then the AI Assistant learns from it and it is able to interact with the user based on this knowledge. 

[taka1]

Hi @igoralcantara,
Thank you for the detailed information.　I really appreciate it.

[igoralcantara]

You got it! 🙂