Project Summary. Neuroimaging provides unparalleled advantages in the ability to understand the neural architecture underlying human thought, feeling, and behavior. Modern approaches combine methodological advances in data acquisition with predictive modeling and larger and more diverse datasets. The result is increasingly sophisticated models that can map patterns of activity onto mental states, behaviors, and experiences. However, there are fundamental limitations impeding progress. First, brains differ in their individual functional topography. Second, it is difficult to make inferences about the spatial topography of brain responses and their variability across individuals. Recent work on inter-subject functional alignment promises to revolutionize brain systems-level modeling of behaviors and mental states by aligning meso-scale activity patterns across individuals. The most popular approach is hyperalignment (HA), which aligns functional brain representations across individuals based on assumptions of a shared representational geometry. However, a large body of work shows that much information about functional brain representations is contained in macro-scale topographical maps. This conventional functional topography is defined on a different, shape-preserving topological manifold better captured using diffeomorphic alignment. We address these issues, providing new ways of modeling topography, making topographical inferences, and making inferences about the topological and geometrical spaces underlying brain representations. We propose to develop diffeomorphic latent space models (DLSMs) that preserve and provide spatial inferences on large-scale topography. Further, we will develop a new class of HA models that place spatial constraints on the transformations, providing fine-scale alignment of representational geometry while minimizing topographical disruption. Finally, we will combine this enhanced HA approach with the DLSM model to create a multi-scale framework that uses diffeomorphic transformations to address large-scale individual differences, followed by geometric transformations to address remaining meso-scale differences. This will allow us for the first time to investigate the relative contributions of topological vs geometrical alignment of data in different brain regions.