Growing neural nets without restarting training
A new method lets you expand a neural network mid-training without disrupting what it already learned.
A new method lets you expand a neural network mid-training without disrupting what it already learned.