A Computational Theory for Sensory Adaptation
Our sensory system consists of multiple processing stages, and its response characteristics change based on recent stimulus history. Although much is known about both the hierarchical nature and the adaptability of the sensory system, it remains unclear how this multilevel neural system adapts to changes in the environment and leads to various perceptual consequences. In my dissertation, I focus on adaptation of the visual motion system. I aim to address the following questions: (I) Can the adaptation of the hierarchical motion system be studied at individual levels of processing? (II) How does the hierarchical motion system adapt and produce perceptual aftereffects? (III) What is the computational principle that underlies multilevel adaptation in the motion system? I designed a novel psychophysical paradigm to study how adaptation of individual levels of motion processing may contribute in producing various forms of perceptual aftereffects. My psychophysical findings suggest that perceptual aftereffects depend on the adaptation of multiple stages of processing, each may adapt based on its own computational principle. Finally, I used a multilayer network model to simultaneously explain different adaptation-induced neural changes that have been observed at different processing levels. Differences in neural adaptation at the local and the global levels of motion processing can be explained from a natural-statistics perceptive: neurons at different levels are holding different assumptions about natural sensory statistics, causing them to react differently in response to the same adapting environment. Taken together, findings from this dissertation illustrate the importance of understanding sensory adaptation from a hierarchical-processing perspective, which will shed more lights on the big picture of sensory processing.