The use of the word ‘inference’ in this context can seem a bit weird, but I think it’s borrowed from statistics and it’s quite standard. Training = optimising

The use of the word ‘inference’ in this context can seem a bit weird, but I think it’s borrowed from statistics and it’s quite standard. Training = optimising model parameters to ‘learn’ from data. Inference = asking the model to make a prediction, usually assuming the model is already trained. Instead of inference, you could say running/querying the model.