Rahul Bordoloi

My research focuses on developing theoretical and statistical machine-learning frameworks for analysing ultrashort time series data and longitudinal data sets. I am particularly interested in deriving probabilistic models that can effectively capture the complex temporal dynamics present in such data while accounting for their sparse and high-dimensional nature. One area of emphasis is extending classical dimension reduction techniques, such as Linear Discriminant Analysis, to handle partially observed ultrashort multivariate time series. This allows projecting the high-dimensional functional data onto a lower-dimensional discriminative subspace, facilitating analysis, interpretation and classification. Additionally, I explore deep generative models that can flexibly learn the underlying distributions governing the evolution of these processes. By combining tools from functional data analysis, time series analysis, and modern machine learning, my work aims to advance the theoretical understanding and practical application of ultrashort time series analysis across diverse domains.