Project Details
Projekt Print View

The role of conditioning in perceptual inference

Subject Area Biological Psychiatry
Human Cognitive and Systems Neuroscience
Term from 2014 to 2018
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 250227358
 
Final Report Year 2018

Final Report Abstract

We investigated the role of learning in perceptual inference, focusing on effects of reward and punishment conditioning on perceptual dominance durations during bistable perception. Related research questions that evolved over the course of the project concerned other forms of associative learning and the role of confidence in learning and inference. We confirmed our initial hypotheses regarding the effects of learning on perceptual inference, as indicated by changes in perceptual dominance durations during bistable perception. Effects of reward and punishment during percept conditioning were in opposite directions, with increased dominance of rewarded percepts and decreased dominance of punished percepts. This finding resembles instrumental conditioning, thus narrowing the gap between the conceptually separated domains of perception and action selection. Interestingly, and in line with our hypotheses, we found that percept conditioning and stimulus conditioning had dissociable effects: In contrast to the observed reduction in dominance durations after aversive percept conditioning, aversive stimulus conditioning yielded an increase in dominance durations, thus suggesting an enhanced salience of the conditioned stimulus. Using fMRI, we could not fully replicate the effects of percept conditioning on perceptual phase durations, but identified metacognitive awareness as an important factor modulating conditioning effects on perceptual inference. The use of a bistable plaid motion stimulus was successful in yielding highly accurate decoding (91% correct) of perceptual states. This is a substantial methodological advance in the field and makes our newly developed paradigm an ideal candidate for objective decoding of subjective perceptual phenomena. Using computational modeling and model-based fMRI, which emerged as an important asset to the project, we found that perceptual inference incorporates rapidly changing expectations from different sources in a manner consistent with Bayesian learning. Model-based fMRI showed that low-level predictions were associated with activity in early visual cortex whereas higher-level predictions correlated with activity in supramodal frontal and temporal regions. Finally, in a model-based fMRI study investigating the role of confidence in learning, we found that subjective confidence involves the mesolimbic reward system and that the neural signals associated with confidence predict its effect on perceptual inference. Taken together, the project yielded important new insights into the role of conditioning and other forms of learning on perceptual inference and the underlying neural mechanisms. Moreover, substantial methodological advances were made with regard to both the development of novel experimental paradigms and the application of advanced computational modelling approaches. The results of this project therefore pave the way for future research into the role of learning processes in perceptual inference in pathological states, such as schizophrenia or depression.

Publications

 
 

Additional Information

Textvergrößerung und Kontrastanpassung