Mixed Linear Regression

Mixed linear regression (MLR) aims to identify multiple underlying linear relationships within a dataset where each data point originates from one of these unknown models. Current research focuses on improving the convergence and robustness of algorithms like Expectation-Maximization (EM) and Alternating Minimization (AM), as well as exploring online and federated learning approaches to handle streaming data and distributed datasets. These advancements address challenges such as imbalanced data, noisy observations, and the need for global convergence guarantees, ultimately enhancing the applicability of MLR in diverse fields requiring the modeling of complex, heterogeneous relationships.