Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!spool.mu.edu!sdd.hp.com!usc!julius.cs.uiuc.edu!psuvax1!news
From: cho@sol4.cs.psu.edu (Sehyeong Cho)
Newsgroups: comp.ai
Subject: info wanted on learning probabilities
Keywords: learning, probability, conditional probabilities
Message-ID: <nj8Gq5ta@cs.psu.edu>
Date: 30 Jan 91 16:42:45 GMT
Sender: news@cs.psu.edu (Usenet)
Organization: Penn State Computer Science
Lines: 19
Nntp-Posting-Host: sol4.cs.psu.edu

Hi, netland.
Has anyone built a system (or read a paper describing a system) that:
 Begins with very rough estimate of cond. prob's.
 Then learns (updates) the conditional probabilities by examples?

For instance, suppose I believed P(death|shoot SCUD) = 0.01,
and then experience a few "shoot SCUD" events along with the
results. (say,  5 SCUD's shot, 1 caused death)
The 5 events are too few to conclude P=0.2. So, it must be
somewhere between 0.01 and 0.2.
Any theory (or heuristics, or psychological evidence..) about how to?

Thanks in advance.

--
                      |  Yesterday I was a student.
     Sehyeong Cho     |  Today I am a student.
     cho@cs.psu.edu   |  Tomorrow I'll probably still be a student.
                      |  Sigh.. There's so little hope for advancement.