A Nonstochastic Control Approach to Optimization
This addresses the challenge of hyperparameter tuning for optimization algorithms, which is incremental as it builds on existing control methods to provide theoretical guarantees.
The paper tackles the nonconvex problem of selecting hyperparameters like learning rate and momentum in optimization by framing it as a meta-optimization problem using online nonstochastic control, and obtains regret guarantees against the best offline solution, ensuring convergence comparable to the best method in hindsight.
Selecting the best hyperparameters for a particular optimization instance, such as the learning rate and momentum, is an important but nonconvex problem. As a result, iterative optimization methods such as hypergradient descent lack global optimality guarantees in general. We propose an online nonstochastic control methodology for mathematical optimization. First, we formalize the setting of meta-optimization, an online learning formulation of learning the best optimization algorithm from a class of methods. The meta-optimization problem over gradient-based methods can be framed as a feedback control problem over the choice of hyperparameters, including the learning rate, momentum, and the preconditioner. Although the original optimal control problem is nonconvex, we show how recent methods from online nonstochastic control using convex relaxations can be used to overcome the challenge of nonconvexity, and obtain regret guarantees against the best offline solution. This guarantees that in meta-optimization, given a sequence of optimization problems, we can learn a method that attains convergence comparable to that of the best optimization method in hindsight from a class of methods.