Inexact bilevel stochastic gradient methods for constrained and unconstrained lower-level problems
This work addresses optimization bottlenecks in machine learning applications like hyperparameter tuning, but it is incremental as it builds on existing bilevel methods with practical improvements.
The paper tackles the challenge of solving large-scale bilevel stochastic optimization problems with nonlinear and nonconvex constraints by introducing new bilevel stochastic gradient methods that avoid second-order derivatives and matrix-vector products, achieving comprehensive convergence theory for both constrained and unconstrained cases.
Two-level stochastic optimization formulations have become instrumental in a number of machine learning contexts such as continual learning, neural architecture search, adversarial learning, and hyperparameter tuning. Practical stochastic bilevel optimization problems become challenging in optimization or learning scenarios where the number of variables is high or there are constraints. In this paper, we introduce a bilevel stochastic gradient method for bilevel problems with nonlinear and possibly nonconvex lower-level constraints. We also present a comprehensive convergence theory that addresses both the lower-level unconstrained and constrained cases and covers all inexact calculations of the adjoint gradient (also called hypergradient), such as the inexact solution of the lower-level problem, inexact computation of the adjoint formula (due to the inexact solution of the adjoint equation or use of a truncated Neumann series), and noisy estimates of the gradients, Hessians, and Jacobians involved. To promote the use of bilevel optimization in large-scale learning, we have developed new low-rank practical bilevel stochastic gradient methods (BSG-N-FD and~BSG-1) that do not require second-order derivatives and, in the lower-level unconstrained case, dismiss any matrix-vector products.