A THREE STATE MARKOV MODEL FOR LEARNING 马尔可夫学习模型的三种状态心理学空间,@b7U*X6CO`

by John Theios心理学空间REi3Bkx#?

$` Q4i5uj;r1k?5}h0TECHNICAL REPORT NO.· 40心理学空间)VE&|5RK&_

`/c(Z[I]f0O0fEv Z0September 22, 1961心理学空间3oe9`b'|O.z

PSYCHOLOGY SERIES心理学空间*OA*Of
|s

G wF0G)E E)Ee0Reproduction in Whole or in Part is Permitted for any Purpose of the United States Government

INSTITUTE FOR· MATHEMATICAL STUDIES IN TRE SOCIAL SCIENCES

&|5I
fY#?:rS#a0RR9B0Applied Mathematics and Statistics Laboratories

!r `^4rtJ0STANFORD UNIVERSITY

sMA|`0n0Stanford, California

&~*|'H{8]
v qI0A THREE STATE MARKOV MODEL FOR LEARNINGY

'A^sb,[1lomEd r0John Theios

0t}n.ik*i!M/y+F%u0Stanford University

D"C!CS*I g7bol0In many learning situations, the response under study initially haszero probability of occurring, but asymptotically, the response probability approaches unity. A few situations of this type are instrumentalavoidance conditioning, classical defense and appetitive conditioning,and reversal learning. These situations will be referred to as simpleconditioning (se). In the past, se has been analyzed in terms of responsestrength or linear operator models (Hull, 1943; Estes, 1950; Bush &Mosteller, 1955), which assume that the strength or probability of alearned response increases gradually during the Course of learning.Recently, it has been found that Markov models, which assume that learning takes place on single trials in an all-or-none fashion, more adequatelydescribe some types of verbal learning than do the linear models (Bower,1960; Estes, 1960). It is quite possible that se is also characterizedby some sort of discrete learning as opposed to gradual learning. Thispossibility is further enhanced by the fact that the "zero to.Unity"response probabilities, characteristic of se, should lend themselvesnicely to discrete conditioning states, which a Markov interpretationwould require (cf., Suppes &Atkinson, 1960). The present paper presentsI am indebted to Dr. Gordon H. Bower for deriving a: number of thetheoretical predictions and for his valuable interest and adviceduring the development of this paper.a three state absorbing Markov model for SC, and then compares thetheoretical predictions to actual data collected in an extensiveexperiment on avoidance conditioning of rats.心理学空间W#H8dNnp$V%R

oJ)dN,lK0全文心理学空间 n\#q&l)sRVH

:G1jOR'h;SQ0http://suppes-corpus.stanford.edu/techreports/IMSSS_40.pdf心理学空间0DV"U}f}