Taming Momentum: Rethinking Optimizer States Through Low-Rank Approximation | ScienceToStartup | ScienceToStartup