Katyusha X: Practical Momentum Method for Stochastic Sum-of-Nonconvex Optimization
This provides an accelerated stochastic algorithm for a core optimization problem in ML, but it is incremental as it builds on the well-known SVRG method.
The paper tackles the problem of minimizing sum-of-nonconvex functions, which is important for tasks like PCA and SVD in machine learning, by introducing a practical momentum method that adds one line to SVRG to achieve acceleration, with results including provable acceleration and linear parallel speed-up using mini-batch.
The problem of minimizing sum-of-nonconvex functions (i.e., convex functions that are average of non-convex ones) is becoming increasingly important in machine learning, and is the core machinery for PCA, SVD, regularized Newton's method, accelerated non-convex optimization, and more. We show how to provably obtain an accelerated stochastic algorithm for minimizing sum-of-nonconvex functions, by $\textit{adding one additional line}$ to the well-known SVRG method. This line corresponds to momentum, and shows how to directly apply momentum to the finite-sum stochastic minimization of sum-of-nonconvex functions. As a side result, our method enjoys linear parallel speed-up using mini-batch.