Modern machine learning applications increasingly involve high-dimensional and complex data, ranging from neuroimaging and genomics to social networks and sensor streams. Such data pose challenges for both computational efficiency and statistical robustness. This project focuses on the development and evaluation of scalable machine learning models tailored to high-dimensional settings. The research will explore a range of machine learning methods, from linear models to deep learning architectures, aiming to balance model expressiveness with computational feasibility. A key component is the design and training of models that generalize well despite the dimensionality, by leveraging data structure and scalable training pipelines. Compute resources from NAISS are essential for this work. The expected outcomes include theoretical advances in machine learning and practical applications in scientific domains that require reliable high-dimensional inference.