The most influential associative learning model, RW1972 (Rescorla & Wagner, 1972), learns from global error and posits no changes in stimulus associability.
Let vk, j denote the associative strength from stimulus k to stimulus j. On any given trial, the expectation of stimulus j, ej, is given by:
$$ \tag{Eq.1} e_j = \sum_{k}^{K}x_k v_{k,j} $$
xk denotes the presence (1) or absence (0) of stimulus k, and the set K represents all stimuli in the design.
Changes to the association from stimulus i to j, vi, j, are given by:
Δvi, j = αiβj(λj − ej)
where αi is the associability of stimulus i, βj is a learning rate parameter determined by the properties of j1, and λj is a the maximum association strength supported by j (the asymptote).
There is no specification of response-generating mechanisms in RW1972. However, the simplest response function that can be adopted is the identity function on stimulus expectations. If so, the responses reflecting the nature of j, rj, are given by:
rj = ej
The implementation of RW1972 allows the specification of
independent β values for
present and absent stimuli (beta_on and
beta_off, respectively).↩︎