The Rescorla-Wagner model is one of the most commonly discussed mathematical models of classical conditioning. It was wildly popular when it came out in 1972, and very successful. The same math, is used in the delta-rule which is a special case of back-propagation with a single layer. In neural network research, the back-propagation algorithm has been refined and replaced over the years. However, for basic classical conditioning, I am not aware of any significant advances over the RW-model. It is still often cited as the model for classical conditioning in textbooks.

However, it is well know that there are numerous conditioning experiments the RW-model cannot explain, for a detailed references, see pages 370 to 378 of Miller et al. (1995). Hence the question:

Is there a modern refinement of the Rescorla-Wagner model that addressed all 23 failures listed in Miller et al. (1995)? How does this refinement relate to refinements of back-propagation?


Miller, R.R., Barnet, R.C., & Grahame, N.J. (1995) "Assessment of the Rescorla-Wagner model." Psychological bulletin 117(3): 363-386 [pdf]


1 Answer 1


I don't think anyone has ever bothered (though Ralph Miller might disagree), since many of the 'failures' are outside of the model's purview. The model expresses as simply as possible the profound insight that we learn most when our expectations are not met. Many features of learning won't conform to this general principle (spontaneous recovery) but it doesn't make it wrong. Other features of learning emerge when the model is applied in layers or a hierarchy, rather than as a single unit (e.g., rapid reacquisition). Hence there is little pressure to adapt the model, although many adaptions have been proposed to account for one effect or another (e.g., Van Hamme & Wasserman, 1994; Dickinson & Burke, 1996 etc). However these have usually limited it's generalisability due to overfitting.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.