I have been reading about "Training and inference" in deep learning. How is this different then the idea of "Train / Test" in general ML? In train/test splits the model has not seen the test set of data. Isn't this the same thing as talking about the 'inference' phase of a deep learning model?

1 Answers

jonnor On Best Solutions

Inference means making predictions using the model. Inference is used during training (the forward pass before back-propagation), and during model evaluation (on the test set), and when the model is used in production (to make predictions on new data).

So when people talk about inference as a phase of development, it typically means in a production setting, after the model has been trained and evaluated (tested).