Abstract
Predictive coding refers to the theory that top-down expectations from higher-level brain areas interact with sensory evidence in lower-level areas. Neural responses in lateral occipital (LO) cortex reflect the consequences of this interaction, with reduced activity when expectations are fulfilled. But where do these predictions come from? We used a multi-session training paradigm and background connectivity to explore the neural mechanisms of prediction based on associations learned either immediately before (“Recent”) or three days before (“Remote”) an fMRI scan. Training sessions were separated in time because different memory systems are thought to support new and old memories. Within each training session, cue stimuli appeared individually and subjects pressed a button to transform the cue into an outcome stimulus. For predictable cues, a particular outcome appeared when the left button was pressed and a different outcome appeared when the right button was pressed. For unpredictable cues, both outcomes appeared with equal probability irrespective of which button was pressed. After training, subjects completed an fMRI session in which Recent and Remote associations were presented in alternating runs. Each run was structured as a block design, with blocks containing either predictable or unpredictable cues. To examine background connectivity during these blocks, we first regressed out variance attributable to stimulus-evoked responses and nuisance variables, and then measured correlations among brain areas in the residual time-series. Correlations were computed in an exploratory manner using the residual time-series in stimulus-selective LO as a predictor of the time-series of all voxels. By comparing predictable to unpredictable blocks, initial results reveal a tradeoff in LO background connectivity for Recent vs. Remote associations between medial temporal lobe structures that support rapid encoding and broader areas of temporal cortex that represent consolidated long-term memories. These findings suggest a neural model of how actions influence predictive coding in the visual system.
Meeting abstract presented at VSS 2015