Proceedings of the Annual Meeting of the Cognitive Science Society

The coarse-to-fine hypothesis posits that, in the Human visualsystem, a coarse representation of visual information is propa-gated quickly through the retina to the cortex, whereas a finer,more detailed representation is propagated more slowly. In aprevious study we showed that recurrent synaptic connectionshelp predict low intensity EFEs. Furthermore, a feedback loopcoming from coarser information processing is postulated toinfluence later processing of finer features. In this paper, weintend to examine the value of coarser information and recur-rence in the processing of dynamic Emotional Facial Expres-sions (EFE). In a step forward in studying the importance ofrecurrent connectivity in the coarse-to-fine model, we testedits advantage for discriminating emotions for different spatialfrequencies and facial expression intensities. Using ArtificialNeural Networks, we modeled recurrent synaptic connectionswith a recurrent feedback loop. Using a Gabor filter bank, wecomputed different levels of spatial frequency features. Our re-sults replicate the advantage of recurrence at first facial expres-sion intensities. Our main finding is that the recurrent model isalso better when predicting high spatial frequencies features.Additionally, mid-to-low spatial frequencies are more usefulto the prediction of EFEs. We conclude that feature process-ing feedback has a significant effect in disambiguating facialexpressions when information is particularly complex, i.e., athigh spatial frequencies and low EFE intensities.