Simulating SCR data based on aversive learning to test different MLE and Bayesian models of Reinforcement Learning models.
Currently I'm presenting to simple models of a task. Task simulate a simple aversive learning task in which participant watches one stimulus at a time. Two total stimuli can be presented, one (CS+) is paired with an electric shock 30% of the time and the other (CS-) is never paired with shock.
The actual task induces a skin-conductance response (SCR) for each of the trials. We expect an elevation in SCR for CS+ and demotion of SCR in CS-.
We use simle Rascorla-Wagner model here which states that:
Expected value = EV[previous trial] + α*PredictionError
When
PE = outcome - expected_value
and
α = subjects' individual learning rate