By speed, several things are meant. This means both classification and learning speed. Symbolic decision trees can obviously be evaluated very quickly, and it is possible to design a tree-builder which builds trees incrementally -- i.e. as new instances come in, they can be added to the tree without having to destroy the old tree. IBLs are also incremental learners. Although naive IBLs can be quite slow, by using techniques like k-d trees quite good performance can be obtained.