Opportunities and challenges in learning to bound

Parameter estimation performance bounds serve as valuable tools in statistical signal processing, yet deriving them traditionally requires full knowledge of the data distribution. Recently, a framework has been proposed that combines a generative model with estimation performance bounds, thus eliminating the need for full knowledge of the data distribution by learning it from data. We refer to this approach as learning-to-bound (L2B). In this paper, we offer a comprehensive review of recent developments and emphasize their advantages. We then dive into open challenges and future directions within the L2B framework. Lastly, we explore a different perspective - the application of estimation performance bounds to deep learning.