Table II.

Comparison of performances on real brain noise test set of LM source localizers with three LM restart strategies, the trained MLP, and a hybrid system

Algorithm	Computation time (ms)	Localization error (cm)
Fixed‐4‐start‐LM	120	0.83
Random‐20‐start‐LM	663	0.54
Optimal‐start‐LM	14	0.49
MLP	0.7	0.90
MLP‐start‐LM	15	0.53

Each number is an average over 25,000 localizations, so error bars are negligible. Training used real brain noise.

LM, Levenberg‐Marquart; MLP, multilayer perceptron