Normalizing Flows: A study on Data Efficient Architectural Design

Understanding astrophysical data often involves analyzing complex distributions where data scarcity can pose significant challenges. To address this, Normalizing Flows offer a promising solution by generating additional data points that align with the underlying distribution. This research project aims to identify which Normalizing Flow architectures are most effective in minimizing the amount of data needed while accurately representing the distribution of interest. By systematically evaluating different flow models, we seek to determine their efficiency in data distribution fitting. This will not only enhance our ability to analyze astrophysical data with limited samples but also provide insights into optimizing flow architectures for various applications in scientific data analysis