Disentangling Adaptive Gradient Methods from Learning Rates
This work clarifies confounding factors in optimization algorithm evaluation for deep learning researchers, though it is incremental in nature.
The paper investigates how adaptive gradient methods interact with learning rate schedules in deep learning, finding that many existing beliefs may stem from insufficient isolation of step size effects through a 'grafting' experiment.
We investigate several confounding factors in the evaluation of optimization algorithms for deep learning. Primarily, we take a deeper look at how adaptive gradient methods interact with the learning rate schedule, a notoriously difficult-to-tune hyperparameter which has dramatic effects on the convergence and generalization of neural network training. We introduce a "grafting" experiment which decouples an update's magnitude from its direction, finding that many existing beliefs in the literature may have arisen from insufficient isolation of the implicit schedule of step sizes. Alongside this contribution, we present some empirical and theoretical retrospectives on the generalization of adaptive gradient methods, aimed at bringing more clarity to this space.