Model training Deep Dive: Refine and maintain (v2024.10)

This course is meant to provide technical audiences and business users with an understanding of the 'Refine' phase of the model training process, as well as how model maintenance works.

It provides insights into precision vs. recall, the role of 'Validation', the model rating and the different performance factors impacting it, and how you can refine your model until it performs as required to meet your objectives. It will also cover how you can maintain your model in production.

The Model training deep dive: Refine and maintain course takes one-hour thirty minutes to complete. At the end, you’ll receive a diploma of completion.

To enjoy this learning experience, we recommend you going through the following courses: 

• UiPath Communications Mining overview 

• Fundamentals of model training 

• Taxonomy design 

• Model training deep dive: Setup 

• Model training deep dive: Discover 

• Model training deep dive: Explore 

The Model training deep dive: Refine and maintain course is aimed at technical audiences and business users, as well as anybody else curious to see the power of Natural Language Processing in business.

The full agenda covers:  

• What is the 'Refine' Phase? 

• Understanding model performance 

• Improving label performance 

• Checking and improving coverage 

• Checking and improving balance 

• Understanding and improving field performance 

• When to stop training your labels and fields 

• How to use 'Train' during the 'Refine' phase 

• Model maintenance 

• Practice: Refine your model in the 'Refine' phase 

At the end of the Model training deep dive: Refine and maintain course, you should be able to: 

• Define what the 'Refine' phase of model training is, how it works, and why it's important. 

• Explain the role of 'Validation' when refining a model. 

• List out the important considerations while labeling during the 'Refine' phase. 

• Explain precision and recall, what a model rating is, and how it's scored. 

• Explain the four factors impacting model performance and how to use the platform's recommended actions. 

• Explain how you can improve the performance of individual labels, and when to use 'Missed Label', 'Check Label', and 'Teach Label' as part of the model refinement process. 

• Explain why coverage is important and how to check and improve your coverage score. 

• Explain why balance is important, and how to check and improve your balance score. 

• Explain how field validation works and when to stop training your fields. 

• Improve the performance of your fields, using the different training modes. 

• Explain when to stop training your labels and fields. 

• Use 'Train' during the 'Refine' phase. 

• Explain how to maintain a model in production, why it's important, and how to add new labels to an existing taxonomy. 

• Prevent concept drift and complete exception training.