Where to grow?

Table 1 Where to grow across algorithms

Algorithm	Where to grow
AutoGrow	Predefined stages.
Firefly	Best loss-improving split or addition.
GradMax	Largest next-step gradient gain.
NeST	Strongest activation-gradient correlations.
Net2Net	All chosen layers at the same time.
Network Morphism	Any chosen layer.
NORTH	Layers lacking novel orthogonal directions.
SENN	Largest residual natural-gradient bottleneck.
Splitting	Neurons with negative splitting criterion.
Tiny	Best residual-gradient match.
Variance Transfer	Underlying width-growth choice.