A neurocognitive framework for understanding how experience shapes object representations

7. 项目总结/文摘

7. PROJECT SUMMARY/ABSTRACT The perceptual system uses prior experience to predict features of incoming stimuli, fill in missing detail, and recognize novel objects based on their relationships to similar, previously experienced objects. Predictions based on prior experience can influence acquisition of object knowledge, causing them to be integrated into representations of previously experienced stimuli. By integrating information acquired across multiple events, people can make novel decisions based on associations that have not been explicitly learned; this ability is thought to be critical for a number of complex behaviors such as semantic learning and spatial navigation. Despite the importance of this integration process, investigations of the neural circuits that shape object representations through predictive mechanisms have only recently begun to examine how information is integrated across separate experiences. Research has demonstrated that the ventral temporal cortex (VTC), hippocampus, and areas of the prefrontal cortex (PFC) are critically involved in integrating new content into existing object representations; however, many questions remain about how these regions work together to combine information from separate learned associations. Theoretical work suggests that integration involves a series of operations, including prediction based on existing associations, detection of an overlap with prior experience, and resolution of interference between competing associations; these processes cannot be separated with simple comparisons based on subsequent behavior. We will use a novel analysis strategy using neuroimaging and computational modeling that will allow us to determine how integration is accomplished in the brain. We will use high-resolution whole-brain functional magnetic resonance imaging (fMRI) to measure activity in regions of the VTC, hippocampus, and PFC, both during learning of associations that overlap with prior experience, and during a task that requires making novel decisions about associations that have not been directly observed. Neural signals measured during learning and testing will be used to constrain the behavior of a computational model of memory integration. Our modeling framework is based on the temporal context model (TCM), which describes operations involved in the construction and maintenance of a temporal context representation that is thought to serve as an overlapping code for bridging between related experiences. The model will be simultaneously constrained by multiple neural measures that provide estimates of variability in each of the computational mechanisms described by the model, allowing us to determine the relationship between different neural signals and the specific computations underlying associative learning. An improved understanding of how prior experience shapes object representations and affects new learning will provide insight into processes that affect perception and comprehension of real-world scenes. Furthermore, the proposed work will develop a neurocognitive modeling framework that will allow construction of personalized models for the memory systems of individual people, making it possible to create more targeted treatments for perceptual and learning deficits.