Continuous-time Models for Stochastic Optimization Algorithms

We propose new continuous-time formulations for first-order stochastic optimization algorithms such as mini-batch gradient descent and variance-reduced methods.