• Is the log-sum-exp trick a valid substitute over scaling when trying to avoid numerical underflow in HMM calculations? http://www.cs.cornell.edu/Courses/cs4758/2012sp/materials/hmm_paper_rabiner.pdf 11 comments 1/11/2012 compsci