Linear Minimization Oracle

Linear Minimization Oracles (LMOs) are fundamental components in optimization algorithms, providing efficient solutions to subproblems within larger optimization tasks. Current research focuses on improving LMO efficiency in various contexts, including stochastic and decentralized optimization, and adapting them for use with non-smooth functions and approximate or "dirty" oracles. These advancements are crucial for scaling optimization methods to high-dimensional problems in machine learning, reinforcement learning, and other fields where computationally expensive exact solutions are impractical.