Machine learning models usually have two stages: the training stage ("Train" in the G2M platform), during which the model learns from the data used in training, and the inference stage ("Infer" in the G2M platform), during which the model generates depending on the use case predictions, insights, etc. These predictions and insights are called inferences. Let's see how it applies to these specific use cases:
Propensity Scoring: The inference stage for propensity scoring models is also known as the Predict stage, since the model is used to generate propensity scores which predict the likelihood an event will happen.
Clustering: The inference stage for clustering models is the stage at which you are using an existing clustering model to assign a cluster ID to a new record. The clusters themselves were determined and identified during the training (fit) stage of the model, and you are now inferring to which cluster a new record should belong.
Regression: similar to propensity scoring, the inference stage for regression models is also known as the predict stage. At that stage you are simply using the model to make predictions using the relationships identified during the training stage.
Media Mix Modeling: during the inference stage of a media mix model you have a model that is past the training stage and you are using the model to generate insights such as optimizing your media mix. You may also be generating predictions though the MMM methodology generally focuses on attribution and optimization insights rather than prediction.
A/B Testing: A/B testing algorithms such as propensity score matching are implemented in a single step in the G2M platform, during the training stage. As such they do not have an Infer stage per se.
Performance Analysis: during the inference stage a performance analysis model will use the knowledge graph and data ingested in training to generate additional insights about any measure of interest.