Single Loop Gaussian Homotopy Method for Non-convex Optimization
About
The Gaussian homotopy (GH) method is a popular approach to finding better stationary points for non-convex optimization problems by gradually reducing a parameter value $t$, which changes the problem to be solved from an almost convex one to the original target one. Existing GH-based methods repeatedly call an iterative optimization solver to find a stationary point every time $t$ is updated, which incurs high computational costs. We propose a novel single loop framework for GH methods (SLGH) that updates the parameter $t$ and the optimization decision variables at the same. Computational complexity analysis is performed on the SLGH algorithm under various situations: either a gradient or gradient-free oracle of a GH function can be obtained for both deterministic and stochastic settings. The convergence rate of SLGH with a tuned hyperparameter becomes consistent with the convergence rate of gradient descent, even though the problem to be solved is gradually changed due to $t$. In numerical experiments, our SLGH algorithms show faster convergence than an existing double loop GH method while outperforming gradient descent-based methods in terms of finding a better solution.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Black-box Adversarial Attack | CIFAR-10 (test) | Success Rate93 | 11 | |
| Global Optimization | Griewank | MSE2.77 | 9 | |
| Global Optimization | Ackley d = 500 | MSE0.57 | 9 | |
| Black-Box Targeted Adversarial Attacks | CIFAR-10 | Success Rate (SR)100 | 9 | |
| Global Optimization | Rosenbrock $d=500$ | MSE0.4 | 9 | |
| Black-box Targeted Adversarial Attack | VitalDB | Success Rate (SR)100 | 9 | |
| Zeroth-order optimization | Theoretical Objective Functions | Iteration Complexity2 | 8 | |
| Adversarial Attack | MNIST 100 images | Success Rate96 | 5 |