home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!warwick!doc.ic.ac.uk!uknet!bcc.ac.uk!link-1.ts.bcc.ac.uk!ubacr45
- From: ubacr45@ucl.ac.uk (Mr G Toal)
- Newsgroups: comp.speech
- Subject: Re: Dynamic Programming
- Keywords: Dynamic Programming, Vitirbi, minimum cost path
- Message-ID: <1992Dec13.050508.24165@bas-a.bcc.ac.uk>
- Date: 13 Dec 92 05:05:08 GMT
- References: <Dec.10.00.19.43.1992.7925@caip.rutgers.edu> <1992Dec11.155402.17484@eng.cam.ac.uk>
- Sender: news@ucl.ac.uk (Usenet News System)
- Organization: Bloomsbury Computing Consortium, London
- Lines: 24
-
- In article <1992Dec11.155402.17484@eng.cam.ac.uk> ajr@eng.cam.ac.uk (Tony Robinson) writes:
- >ergodic Markov model". To take away some of the jargon, I want to recognise
- >a sentence as a sequence of phones. This means allocating one phone to each
- >time slot. This could be done by generating all possible sequences,
- >computing a cost for each one, and picking the one with the least cost.
- >This is horrendously inefficient, and that is where dynamic programming
- >comes in. By computing an optimal cost to a partial solution, say from the
- >start to time t, then to find an optimal solution to time t+1 you only have
- >to look at the solutions to time t and all the possible extensions. This is
- >the best I can put it in a reasonable amount of time, but it it this
- >"principle of optimality" that it is important to understand. Dynamic
- >programming makes the problem linear in time, as opposed to exponential - a
- >huge win!
-
- I've heard the phrase 'dynamic time warp' bandied around, but your
- description here of dynamic programming is what I thought was dynamic
- time warp. (I'd been under the impression that dynamic programming
- was what we used to call in the seventies in AI 'memo functions')
-
- If you know about dynamic time warp too, Tony, could I persuade you
- to jot down a few lines like you did for this (for which edification
- I thank you, by the way.)
-
- Graham
-