- for each training example, randomly delete nodes in the [[Deep Learning|Neural Network]], & for the back propagation tuning of weights, you follow this reduced network for that example ![[CleanShot 2024-06-13 at [email protected]|300]] - during test time, you won't use dropout anymore