This (assuming the phenomenon is real) seems hard to explain in the way that intelligence is normally considered - in terms of intelligence being a consequence of very large numbers (thousands?) of genes-for-intelligence. With intelligence genes conceptualized as additive in effect, and in such large numbers, it is hard to understand how a very highly intelligent child could emerge by chance from low intelligence parents.
But if person's level of intelligence is also determined by the number of deleterious mutations they inherit from their parents, and these mutations are numbered in tens - then it is imaginable that, by chance, a child may be born with very few deleterious mutations, despite his parents having a relatively heavy mutation load.
This notion is perhaps testable, on the basis that a low mutation load should be associated with generally higher fitness - so the high intelligence child of low intelligence parents would be expected to be (on average) taller, healthier, more symmetrical, more long-lived than his low intelligence parents.