Documente Academic
Documente Profesional
Documente Cultură
33 and the
probability of B = 0.2.
Rescorla – Wagner Model for multiple stimuli is given by:
Wi+1 = Wi + ε. Δi.ui; ……. (1)
Δi = ri - vi ; ………………. (2)
Wss is the asymptotic weights obtained at steady state, the average prediction error
vanishes:
Wss = Q-1 . < r.u >, where Q = <u.u> …………. (3)
Where Q-1 is the inverse of the autocorrelation matrix Q (which exists for
sufficiently decorrelated elements of u).
ui – matrix indicating presence or absence of 2 stimuli .
r – vector indicating reward awarded or withheld
Wi – weight vector of each trial.
Task 1 :
The probabilities of stimulus A and B are shown in table below :
The probability table is given by:
Task 2:
We are performing complete reinforcement. In which independent stimulus
probabilities is given by 1/3 and 1/5 of stimulus A and B respectively, reward
association <r|u> = [1; 0.2667] and Wss = [1; 0].
Both are not equal there are rewards accidentally traced back to stimulus B , although
b does not predict any reward over and above A. The graph results are shown in figure
1. The weights tends to be 1 and the Δ which is the difference between expected and
real reward tends to 0.
Task 3:
Figure 2 : The development of reward expectations over time (trials)
As shown in figure 2 , it depicts that specific reward expectation w1 foe stimulus A
converges to 0.5, that is the probability for getting a reward when stimulus A is
present pr =0.5 .The specific reward expectation w2 for stimulus B converges to 0.
Task 4:
Figure 3 shows that the specific reward expectation 𝑤1 for stimulus A converges to
approximately 0.8. The specific reward expectation 𝑤2 for stimulus B converges to
approximately −0.1.
A present stimulus B when A is present is a punishment.