Abstract
Does perceptual learning (PL) follow the same rule as classical conditioning? In classical conditioning, it is suggested that conditioning only occurs when the relative probability of occurrence of unconditioned stimulus (US) in the presence of conditioned stimulus (CS) differs from that in the absence of CS. When the probability of US is higher during CS than at other times, excitatory conditioning occurs and when the probability is lower, inhibitory conditioning occurs. Last year we showed that PL in humans occurs for stimuli that are consistently paired with a liquid reward (Kim, Seitz, Watanabe, VSS, 2007). This raises the possibility that the common mechanism underlies between classical conditioning and PL shaped by reward. To test this hypothesis, we have examined how learning occurs as a function of reward-contingency. Using a classical conditioning paradigm, we presented every 500 msecs a different noise image that filled the display. At random intervals, sinusoidal gratings (20% signal; 2 c/deg; 4 deg diameter) were superimposed on the noise background. To vary the probability of reward-delivery three different orientations were used for each subject: (a) the zero-contingency orientation had reward-probability equal to the background reward-rate of 50%, (b) the positive-contingency orientation had probability of reward of 80%, and (c) the negative-contingency orientation had probability of reward of 20%. Sensitivity tests for the three orientations were conducted before and after nine days of conditioning. We found significant performance improvement for the positive-contingency orientation, marginal improvement for the negative-contingency orientation and no significant improvement for the zero-contingency orientation. These results demonstrate that reward contingencies (not just stimulus reward pairings) shape perceptual learning. However, the performance improvement observed for the inhibitory orientation suggests that perceptual learning does not follow the same rules as classical conditioning.
This study is funded by National Institutes of Health (R21 EY017737), the Human Frontier Science Program Organization (RGP18/2004).