Inference

Using a trained model to process new inputs and create outputs. This is the operational phase, as distinguished from training. Inference is distinguished from training by the way in which data is handled: training uses input data to adjust model weights, leading to possible issues with memorization, while inference only uses input data transiently to produce outputs. Inference is also distinguished from generation; a classification system performs inference, but the scope of possible outputs is limited. In contrast, generative systems may reproduce input or training data. Contracts often define "inference" and "training" separately and impose different restrictions on each.