Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Limited-Memory Matrix Adaptation for Large Scale Black-box Optimization

About

The Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is a popular method to deal with nonconvex and/or stochastic optimization problems when the gradient information is not available. Being based on the CMA-ES, the recently proposed Matrix Adaptation Evolution Strategy (MA-ES) provides a rather surprising result that the covariance matrix and all associated operations (e.g., potentially unstable eigendecomposition) can be replaced in the CMA-ES by a updated transformation matrix without any loss of performance. In order to further simplify MA-ES and reduce its $\mathcal{O}\big(n^2\big)$ time and storage complexity to $\mathcal{O}\big(n\log(n)\big)$, we present the Limited-Memory Matrix Adaptation Evolution Strategy (LM-MA-ES) for efficient zeroth order large-scale optimization. The algorithm demonstrates state-of-the-art performance on a set of established large-scale benchmarks. We explore the algorithm on the problem of generating adversarial inputs for a (non-smooth) random forest classifier, demonstrating a surprising vulnerability of the classifier.

Ilya Loshchilov, Tobias Glasmachers, Hans-Georg Beyer• 2017

Related benchmarks

TaskDatasetResultRank
High-dimensional optimizationMSLR
Convergence Value-8.6829
21
High-dimensional optimizationLasso-Hard
Convergence Value5.5371
20
High-dimensional optimizationLIMO
Convergence Value-6.5043
20
Function OptimizationRosenbrock D=1000
Convergence Value4.48e+5
19
Function OptimizationLevy D=1000
Convergence Value98.4986
19
Function OptimizationSphere D=1000
Final Value99.1157
19
Function OptimizationDixon D=1000
Convergence Value6.82e+5
19
Function OptimizationGriewank D=1000
Convergence Value (Statistic)83.7417
19
Function OptimizationMichalewicz D=1000
Convergence Value-7.3345
19
High-dimensional optimizationRosenbrock D=10000
Convergence Value5.72e+5
13
Showing 10 of 15 rows

Other info

Follow for update