Linear Viterbi Algorithm

1
Linear Viterbi Algorithm

• ?t1(j) b j(o t1)max(?t(i)aij)
• If aij e-ki-j, then
• ?t1 (j) b j (o t1) min(?t (i) aij )
• D(j) min ?(i - j) f(i)
• To compute D(j), we only need a two-pass
  algorithm, which is O(n), instead O(n2).

2
Application

• A coin-tossing model with a one-dimensional
  embedding of states to estimate the download
  probability of items at a web site.
• This Application is used to tracking and
  measuring the changes in user interest in an item.