Abstract: Brain connectomics has become a central tool in neuroimaging for advancing our understanding of neural circuits and their roles in neurodevelopment, mental illness, and aging. However, these analyses face major challenges, including the high dimensionality of brain networks, the presence of latent sources underlying observed connectivity, and the large number of connections that can lead to spurious findings. In this talk, we introduce LOCUS, a regularized blind source separation (BSS) framework for reliable mapping of neural circuits from static and dynamic functional connectomes, as well as for the integrative analysis of multi-view connectomes across imaging modalities and cognitive states. The proposed methods leverage low-rank factorization, a novel angle-based sparsity control, and regularizations to achieve efficient and robust source separation of connectivity matrices. We develop highly efficient algorithms to solve the non-convex optimization problem for learning the proposed models. Applications to large-scale neuroimaging studies demonstrate substantially improved reproducibility in identifying neural circuits and their associations with demographic and clinical phenotypes. The findings provide new insights into brain network organization, dynamics and adaptations.