The 4 Models of AI Builder
With AI Builder, you have 4 categorisations of the available maching learning models that you can build.
Prediction
A Prediction model takes your previous data in order to make an estimate in the future, usually using historical patterns to determine whether or not something will occur in the future. Essentially, by taking historical data and the outcome of the event, the AI can make a yes/no prediction as a future outcome.
Coin-flip prediction example If you feed an AI data from 50 coin flips, it will try to make a guess on the outcome the next coin flip!
Now in terms of business, there are some unique problems that could utilise AI to help predict important vectors to accelerate the growth of a business. These could entail:
Whether a lead will become a customer or not
Will a project become profitable (based on previous project history)
Will a customer churn based on their recent activity?
i.e. will a customer stop using the services / cancel their subscription?
some indicators could be less frequent use or even a spike in usage before the renewal date
Form Processing
For Form Processing, the AI Model deals with taking text from objects (could be in an image, photograph or real-time) and extracting the desired data. For example, the business card reader is a type of form processing, where the text information is queried and found on the card based on structured patterns. The AI could be looking for things like area codes in numbers, or street address indicators, etc.
Furthermore, this could be extended for real-time form processing, taking data from reality (such as Google Lens does) and interpreting the text in your environment. From there, the text data that is extracted can be useful to perform actions like searching with the data, or translating the text, etc.
Object Detection
Object Detection in AI Builder mainly deals with identifying certain objects when given a picture. By analysing a photo, the AI model looks for key indicators that can help to determine whether an object exists in the frame or not. The user having trained the AI by giving it a set of labelled data provides the information necessary for the AI to recognise patterns (such as two wheels indicating a bicycle, or the shape of a helmet).
One thing that's worth mentioning is with object detection, without a good variation in data, the AI Model may infer a certain level of bias affecting its accuracy with different images. For example, if you train a model with photos of bicycles from a side view, it most likely will have a hard time trying to identify a bicycle in a picture if it is in a different orientation. For this reason, training an object detection model requires a minimum number of images for training data as well as image data the object under different angles and lighting conditions.
Text Classification
Finally, text classification takes text and attempts to infer its meaning for analysis. These classifiers assign tags to text and categorise the data it is given based on the content of the text. This is useful where you have a set of text that is unstructured and needs grouping to make analysis on it easier.
For example, text classification can be used for things like:
Sentiment analysis - determining whether the context of a text block is positive / negative
Topic labeling - grouping text information under a certain topic
Intent detection - determines the intent of the customer based on their conversations
Last updated