Recent advances in neuroimaging techniques have enabled researchers to investigate the human brain at unprecedented levels of detail. In particular, functional magnetic resonance imaging (fMRI) has become a powerful tool for studying brain activity and connectivity. However, the complexity of fMRI data poses significant challenges for analysis and interpretation, requiring the development of novel computational approaches. This thesis aims to contribute to this field by investigating two key aspects related to fMRI analysis, connectivity and activity, drawing on insights from machine learning (ML), computer vision, and neuroscience. The first part of this thesis focuses on fMRI connectivity analysis. It is a common assumption about the brain that white matter structural connections are likely to support flow of functional activation or functional connectivity. While the relationship between structural (SC) and functional connectivity (FC) profiles, here called SC-FC coupling, has been studied on a whole-brain, global level, few studies have investigated this relationship at a regional scale. In this part, we quantify regional SC-FC coupling in healthy young adults using diffusion-weighted MRI and resting-state functional MRI data from the Human Connectome Project (HCP) and study how SC-FC coupling may be heritable and vary between individuals. We show that regional SC-FC coupling strength varies widely across brain regions, but was strongest in highly structurally connected visual and subcortical areas. We also show inter-individual regional differences based on age, sex and composite cognitive scores, and that SC-FC coupling is highly heritable within certain networks. Our results suggest regional structure-function coupling is an idiosyncratic feature of brain organisation that may be influenced by genetic factors. The second part of this thesis aims to investigate human brain's regional activation selectivity and inter-individual differences in human brain responses to various visual stimuli. Building computational encoding models that map images to neural responses is one way to pursue this goal. Moreover, generating or selecting visual stimuli designed to achieve specific patterns of responses allows exploration and control of neuronal firing rates or regional brain activity responses. Towards this end, we propose a computational strategy framework, called NeuroGen, to combine the neural encoding model with state-of-the-art generative model and produce high fidelity images that can achieve targeted brain activation patterns. We first show that NeuroGen can serve as a robust discovery architecture for visual neuroscience, including differences in regional and individual human brain response patterns to visual stimuli. We next explore different personalized encoding model architectures, and propose an ensemble approach that has the best balance between model accuracy and the ability to preserve patterns of inter-individual differences in the image-response relationship, to be plugged in NeuroGen for individual level image synthesis. The NeuroGen framework is validated with two fMRI experiments, where we show the selected natural images and generated synthetic images to new subjects and collected their brain responses to these visual stimuli. The results demonstrate that data-driven and generative modeling framework can be leveraged to probe inter-individual differences in and functional specialization of the human visual system. And for the first time, we indicate that NeuroGen can be used to modulate macro-scale brain regions in specific individuals using synthetically generated visual stimuli. Finally, the last section of the second part presents a surface-based convolutional network for reconstructing natural image stimuli from fMRI data. We show that taking advantage of the spatial organization of the brain's cortical surface can improve the accuracy of decoding, and achieve state-of-the-art performance. Overall, the work presented in this thesis contributes to the field of neuroscience by advancing our understanding of the neural mechanisms underlying perception and cognition. Specifically, it demonstrates the importance of regional structure-function coupling of the brain, and highlights the potential of ML techniques for understanding the relationship between external stimuli and human brain responses. These findings have implications for a wide range of fields, from fundamental neuroscience research to clinical applications such as neuroimaging-based diagnosis and treatment of neurological disorders. This thesis underscores the importance of interdisciplinary collaborations between neuroscience and ML, and provides a foundation for further advancements in this exciting and rapidly evolving field.