Custom models
The Refine a model for your business needs section of the AI Builder’s Build screen contains tiles that provide access to models that developers can customize for use with their own data and business needs. Unlike the prebuilt models, which use AI to make determinations based on standard business practices, developers must build and train the custom models with their own data and practices. For example, anyone can use the prebuilt Business Card Reader model without modification of the AI, because the information found on business cards is predictable. For custom models, such as Object Detection, the AI must be trained to recognize images of an organization’s products before it can identify them in images accurately.
The custom models provided by AI Builder include the following:
■ Category Classification—Analyzes text and assigns tags to it representing categories of any type
■ Entity Extraction—Scans incoming text for specific data elements and uses those elements to categorize the text
■ Form Processing—Scans forms and extracts the pertinent data for storage or use by the flow or app
■ Object Detection—Scans image files for recognized objects and identifies them
■ Prediction—Uses historical patterns of past results to predict future responses in binary solution sets, such as yes/no, true/false, and go/no go
The process of building and training one of the custom models requires the developer to supply training data so that the AI can learn to perform the required task. For example, to use the Category Classification model, the developer must supply samples of data that has already been tagged categorically and a minimum of 10 examples for each category tag. Using this information, the AI can learn what kind of data each tag represents and prepare itself to duplicate the tagging process on new data supplied to it. In the same way, the developer must supply multiple samples of the same forms for the Form Processing model and images of the objects to be identified for the Object Detection model.
Some of the AI Builder models, such as Category Classification and Entity Extraction, are available both as prebuilt and custom models. The primary difference between the two versions is that the prebuilt models have AI functions that are already trained with generic data, whereas the custom models need to be trained with data supplied by the developer. If the AI function in the prebuilt model is satisfactory for a developer’s particular application, it is possible to customize the triggers that launch the flow or the actions performed after the AI process is completed. To modify the functionality of the AI, the developer must use the custom model and train it with new data.