The paper studies nonsmooth contractive stochastic approximation with a constant step size, motivated by Q-learning. It examines both additive-noise dynamics and synchronous as well as asynchronous variants, establishing steady-state convergence through a prelimit coupling method. The work is an updated arXiv submission.
This paper examines continuous-time dropout within controlled differential equations by approximating additive vector fields with random batches. Over each interval of length h, a random subset of components is activated and rescaled according to its inclusion probability. The work covers convergence, design, and training aspects of the approach.
The paper compares two federated learning setups: parallel training, where clients work simultaneously, and sequential training, where clients update the model one after another. It derives convergence guarantees for the sequential approach when client data distributions differ, a setting where parallel methods often struggle. The analysis aims to clarify when sequential federated learning offers theoretical advantages over its parallel counterpart.