Lukás Bajer

Benchmarking Gaussian Processes and Random Forests Surrogate Models on the BBOB Noiseless Testbed

Speeding-up black-box optimization algorithms via learning and using a surrogate model is a heavily studied topic. This paper evaluates two different surrogate models: Gaussian processes and random forests which are interconnected with the state-of-the art optimization algorithm CMA-ES. Results on the BBOB testing set show that considerable amount of fitness evaluations can be saved especially during the initial phase of the algorithms progress.

Neural networks as surrogate models for measurements in optimization algorithms

The paper deals with surrogate modelling, a modern approach to the optimization of objective functions evaluated via measurements. The approach leads to a substantial decrease of time and costs of evaluation of the objective function, a property that is particularly attractive in evolutionary optimization. The paper recalls common strategies for using surrogate models in evolutionary optimization, and proposes two extensions to those strategies - extension to boosted surrogate models and extension to using a set of models. These are currently being implemented, in connection with surrogate modelling based on feed-forward neural networks, in a software tool for problem-tailored evolutionary optimization of catalytic materials. The paper presents results of experimentally testing already implemented parts and comparing boosted surrogate models with models without boosting, which clearly confirms the usefulness of both proposed extensions.

Surrogate modeling in the evolutionary optimization of catalytic materials

The search for best performing catalysts leads to high-dimensional optimization tasks. They are by far most frequently tackled using evolutionary algorithms, usually implemented in systems developed specifically for the area of catalysis. Their fitness functions are black-box functions with costly and time-consuming empirical evaluation. This suggests to apply surrogate modeling. The paper points out three difficulties challenging the application of surrogate modeling to catalysts optimization: mixed-variables optimization, assessing the suitability of different models, and scalarization of multiple objectives. It then provides examples of how those challenges are tackled in real-world catalysts optimization tasks. The examples are based on results obtained in three such tasks using one of the leading specific evolutionary optimization systems for catalysis.

Investigation of Gaussian Processes and Random Forests as Surrogate Models for Evolutionary Black-Box Optimization

This paper introduces two surrogate models for continous black-box optimization, Gaussian processes and random forests, as an alternative to the already used ordinal SVM regression. We employ the CMA-ES as the reference optimization method with which the surrogate models are combined and also compared on subset of the noisless BBOB testing set.

Case study: constraint handling in evolutionary optimization of catalytic materials

The paper presents a case study in an industrially important application domain the optimization of catalytic materials. Though evolutionary algorithms are the by far most frequent approach to optimization tasks in that domain, they are challenged by mixing continuous and discrete variables, and especially by a large number of constraints. The paper describes the various kinds of encountered constraints, and explains constraint handling in GENACAT, one of evolutionary optimization systems developed specifically for catalyst optimization. In particular, it is shown that the interplay between cardinality constraints and linear equality and inequality constraints allows GENACAT to efficienlty determine the set of feasible solutions, and to split the original optimization task into a sequence of discrete and continuous optimization. Finally, the genetic operations employed in the discrete optimization are sketched, among which crossover is based on an assumption about the importance of the choice of sets of continuous variables in the cardinality constraints.