This sounds like an expectation-maximization approach to multi-mapping reads. I am not aware of any aligner doing this, though EM algorithms are common when counting multi-mapped reads (e.g. RSEM).

What is common among aligners is to randomly choose one location among the many equally good possible locations, but this is simpler than an EM algorithm.

edit: I should have googled before posting, here is a pre-print about EM-mapping.