Search Papers | Poster Sessions | All Posters
Poster A101 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Divisive normalization captures perceptual and neural interaction effects between temporal adaptation and contrast gain during object recognition
Amber M. Brands1 (), Zilan Oz1, Iris I.A. Groen1,2; 1Informatics Institute, University of Amsterdam, 2Department of Psychology, University of Amsterdam
Human perception remains robust under challenging viewing conditions. This robustness in perception has been linked to nonlinear processing of visual inputs. Here, we combine human EEG, behavior and deep neural network modeling to examine the joint impact of two nonlinear response properties, namely temporal adaptation and contrast gain, on perception of objects embedded in temporally repeated noise. We observe an interaction effect, with higher categorization performance when adapting to noise for high, but not lower object contrast levels. This improved performance is associated with more pronounced contrast-dependent modulation of the evoked neural responses and enhanced decoding of object identity. Using deep convolutional neural networks, we demonstrate that interaction effects between temporal adaptation and contrast level are effectively captured by temporal divisive normalization. Moreover, examining the network representations reveals that, similar to the neural data, adapting to the same noise results in improved representations of the object due to noise suppression. Overall, our findings shed light on how benefits of temporal adaptation are influenced by contrast level and offer an intuitive framework to study the integration of nonlinear response properties and their impact on perception.
Keywords: temporal adaptation contrast gain object recognition divisive normalization