Abstract
Bayesian Optimization has become a widely used approach to perform optimization involving computationally intensive black-box functions, such as the design optimization of complex engineering systems. It is often based on Gaussian Process regression as a Bayesian surrogate model of the exact functions. Bayesian Optimization has been applied to single and multi-objective optimization problems. In case of multi-objective optimization, the Bayesian models used in optimization often consider the multiple objectives separately and do not take into account the possible correlation between them near the Pareto front. In this paper, a Multi-Objective Bayesian Optimization algorithm based on Deep Gaussian Process is proposed in order to jointly model the objective functions. It allows to take advantage of the correlations (linear and non-linear) between the objectives in order to improve the search space exploration and speed up the convergence to the Pareto front. The proposed algorithm is compared to classical Bayesian Optimization in four analytical functions and two aerospace engineering problems.
Similar content being viewed by others
Data Availibility Statement
The datasets and analytical problems generated and supporting the findings of this article, are obtainable from the corresponding author upon reasonable request.
References
Abadi M, Barham P, Chen J, Chen Z, Davis A, Dean J, Devin M, Ghemawat S, Irving G, Isard M, et al. (2016) Tensorflow: a system for large-scale machine learning. In: 12th USENIX symposium on operating systems design and implementation (OSDI 16), pp 265–283
Alvarez MA, Rosasco L, Lawrence ND (2011) Kernels for vector-valued functions: a review. arXiv:1106.6251
Arias-Montano A, Coello CAC, Mezura-Montes E (2012) Multiobjective evolutionary algorithms in aeronautical and aerospace engineering. IEEE Trans Evolut Comput 16:662–694
Auger A, Bader J, Brockhoff D, Zitzler E (2012) Hypervolume-based multiobjective optimization: theoretical foundations and practical implications. Theoret Comput Sci 425:75–103
Bader J, Zitzler E (2011) HypE: an algorithm for fast hypervolume-based many-objective optimization. Evol Comput 19:45–76
Beume N, Naujoks B, Emmerich M (2007) SMS-EMOA: multiobjective selection based on dominated hypervolume. Eur J Oper Res 181:1653–1669
Bradstreet L (2011) The hypervolume indicator for multi-objective optimisation: calculation and use. Ph.D. thesis, University of Western Australia Perth
Brevault L, Balesdent M, Hebbal A (2020) Multi-objective multidisciplinary design optimization approach for partially reusable launch vehicle design. J Spacec Rockets 58:1–17
Bui T, Hernández-Lobato D, Hernandez-Lobato J, Li Y, Turner R (2016) Deep Gaussian processes for regression using approximate expectation propagation. In: International conference on machine learning, pp 1472–1481
Bussemaker JH, Ciampa PD, Nagel B (2020) System architecture design space exploration: an approach to modeling and optimization. AIAA Aviation 2020 Forum. 3172
Castellini F, Riccardi A, Lavagna M, Büskens C (2011) Global and local multidisciplinary design optimization of expendable launch vehicles. In: 52nd AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics and materials conference
Chauhan SS, Martins JRRA (2018) Low-fidelity aerostructural optimization of aircraft wings with a simplified wingbox model using OpenAeroStruct. In: International conference on engineering optimization, pp 418–431
Chiplunkar A, Rachelson E, Colombo M, Morlier J (2016) Approximate inference in related multi-output Gaussian process regression. In: International conference on pattern recognition applications and methods, pp 88–103
Constantinescu EM, Anitescu M (2013) Physics-based covariance models for Gaussian processes with multiple outputs. Int J Uncertain Quantif 3(1):47–71
Cutajar K, Bonilla EV, Michiardi P, Filippone M (2017) Random feature expansions for deep Gaussian processes. In: International conference on machine learning, pp 884–893
Cutajar K, Pullin M, Damianou A, Lawrence N, González J (2019) Deep Gaussian processes for multi-fidelity modeling. arXiv:1903.07320
Dai Z, Damianou A, González J, Lawrence N (2015) Variational auto-encoded deep Gaussian processes. arXiv:1511.06455
Damianou A (2015) Deep Gaussian processes and variational propagation of uncertainty. Ph.D. thesis, University of Sheffield
Damianou A, Lawrence N (2013) Deep Gaussian processes. Artif Intell Stat 3(1):47–71
De Matthews AG, Van Der Wilk M, Nickson T, Fujii K, Boukouvalas A, León-Villagrá P, Ghahramani Z, Hensman J (2017) GPflow: a Gaussian process library using TensorFlow. J Mach Learn Res 18:1299–1304
Deb K (2001) Multi-objective optimization using evolutionary algorithms, vol 16. Wiley, New York
Deb K, Agrawal S, Pratap A, Meyarivan T (2000) A fast elitist non-dominated sorting genetic algorithm for multi-objective optimization: NSGA-II. In: International conference on parallel problem solving from nature, pp 849–858
Deb K, Thiele L, Laumanns M, Zitzler E (2005) Scalable test problems for evolutionary multiobjective optimization. Evolutionary multiobjective optimization. Springer, Berlin, pp 105–145
Durantin C, Marzat J, Balesdent M (2016) Analysis of multi-objective Kriging-based methods for constrained global optimization. Comput Optim Appl 63:903–926
Emmerich M, Klinkenberg J (2008) The computation of the expected improvement in dominated hypervolume of Pareto front approximations. Rapport Techn Leiden Univ 34:3–7
Emmerich M, Giannakoglou KC, Naujoks B (2006) Single-and multiobjective evolutionary optimization assisted by Gaussian random field metamodels. IEEE Trans Evolut Comput 10:421–439
Emmerich M, Yang K, Deutz A, Wang H, Fonseca CM (2016) A multicriteria generalization of Bayesian global optimization. In: Advances in stochastic and deterministic global optimization. Springer, pp 229–242
Forrester A, Sobester A, Keane A (2008) Engineering design via surrogate modelling: a practical guide. Wiley, New York
Geman S, Geman D (1984) Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. In: IEEE transactions on pattern analysis and machine intelligence PAMI-6, pp 721–741
Hadka D (2015) Platypus-multiobjective optimization in python
Haibin Y, Chen Y, Low BKH, Jaillet P, Dai Z (2019) Implicit posterior variational inference for deep Gaussian processes. In: Advances in neural information processing systems, pp 14502–14513
Hebbal A (2021) Deep Gaussian processes for the analysis and optimization of complex systems-application to aerospace system design. Ph.D. thesis, Université de Lille
Hebbal A, Brevault L, Balesdent M, Talbi E-G, Melab N (2019a) Bayesian optimization using deep Gaussian processes. arXiv:1905.03350
Hebbal A, Brevault L, Balesdent M, Talbi E-G, Melab N (2019b) Multi-fidelity modeling using DGPs: improvements and a generalization to varying input space dimensions. Conference: 33rd conference on neural information processing systems, Vancouver
Hernández-Lobato D, Hernandez-Lobato J, Shah A, Adams R (2016) Predictive entropy search for multi-objective Bayesian optimization. In: International conference on machine learning, pp 1492–1501
Jasa JP, Hwang JT, Martins JRRA (2018) Open-source coupled aerostructural optimization using Python. Struct Multidiscip Optimiz 57:1815–1827
Jeong S, Obayashi S (2005) Efficient global optimization (EGO) for multi-objective problem and data mining. In: 2005 IEEE congress on evolutionary computation, pp 2138–2145
Jones DR, Schonlau M, Welch WJ (1998) Efficient global optimization of expensive black-box functions. J Glob Optim 13:455–492
Knowles J (2006) ParEGO: a hybrid algorithm with on-line landscape approximation for expensive multiobjective optimization problems. IEEE Trans Evol Comput 10:50–66
Kullback S (1997) Information theory and statistics. Courier Corporation, North Chelmsford
Kursawe F (1990) A variant of evolution strategies for vector optimization. In: International conference on parallel problem solving from nature, pp 193–197
Liu W, Zhang Q, Tsang E, Liu C, Virginas B (2007) On the performance of metamodel assisted MOEA/D. In: International symposium on intelligence computation and applications, pp 547–557
Matheron G (1967) Kriging or polynomial interpolation procedures. CIMM Trans 70:240–244
Nebro AJ, Durillo JJ, Garcia-Nieto J, Coello CAC, Luna F, Alba E (2009) SMPSO: a new PSO-based metaheuristic for multi-objective optimization. In: 2009 IEEE symposium on computational intelligence in multi-criteria decision-making, pp 66–73
Oliver MA, Webster R (1990) Kriging: a method of interpolation for geographical information systems. Int J Geograph Inf Syst 4:313–332
Perdikaris P, Raissi M, Damianou A, Lawrence ND, Karniadakis GE (2017) Nonlinear information fusion algorithms for data-efficient multi-fidelity modelling. Proc Roy Soc A Math Phys Eng Sci 473:20160751
Picheny V (2015) Multiobjective optimization using Gaussian process emulators via stepwise uncertainty reduction. Stat Comput 25:1265–1280
Ponweiser W, Wagner T, Biermann D, Vincze M (2008) Multiobjective optimization on a limited budget of evaluations using model-assisted S-metric selection. In: International conference on parallel problem solving from nature, pp 784–794
Salimbeni H, Deisenroth M (2017) Doubly stochastic variational inference for deep Gaussian processes. In: Advances in neural information processing systems, pp 4588–4599
Schonlau M, Welch WJ, Jones D (1996) Global optimization with nonparametric function fitting. In: Proceedings of the ASA, section on physical and engineering sciences, pp 183–186
Shah A, Ghahramani Z (2016) Pareto frontier learning with expensive correlated objectives. In: International conference on machine learning, pp 1919–1927
Stein ML (1999) Interpolation of spatial data: some theory for kriging. Springer, Berlin
Svenson JD, Santner TJ (2016) Multiobjective optimization of expensive-to-evaluate deterministic computer simulator models. Comput Stat Data Anal 94:250–264
Talbi E-G (2009) Metaheuristics: from design to implementation, vol 74. Wiley, New York
Talbi E-G, Basseur M, Nebro AJ, Alba E (2012) Multi-objective optimization using metaheuristics: non-standard algorithms. Int Trans Oper Res 19:283–305
Toal DJJ, Keane AJ (2012) Non-stationary kriging for design optimization. Eng Optim 44:741–765
Wang Y, Brubaker M, Chaib-Draa B, Urtasun R (2016) Proceedings of the 19th International conference on artificial intelligence and statistics (AISTATS), vol. 51. JMLR: W&CP, Cadiz, Spain
Williams CKI, Rasmussen CE (2006) Gaussian processes for machine learning, vol 2. MIT Press, Cambridge
Yang K, Emmerich M, Deutz A, Bäck T (2019) Efficient computation of expected hypervolume improvement using box decomposition algorithms. J Global Optim 75:3–34
Zhang Q, Liu W, Tsang E, Virginas B (2009) Expensive multiobjective optimization by MOEA/D with Gaussian process model. IEEE Trans Evol Comput 14:456–474
Zhang Q, Liu W, Tsang E, Virginas B (2010) Expensive multiobjective optimization by MOEA/D with Gaussian process model. IEEE Trans Evol Comput 14:456–474
Zitzler E, Laumanns M, Thiele L, Fonseca CM, da Fonseca VG (2002) Why quality assessment of multiobjective optimizers is difficult. In: Proceedings of the 4th annual conference on genetic and evolutionary computation, pp 666–674
Acknowledgements
The work of Ali Hebbal is part of a PhD thesis co-funded by ONERA—The French Aerospace Lab and the university of Lille.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix: numerical setup
Appendix: numerical setup
-
The Experiments presented in this paper were carried out using the Grid’5000 testbed, supported by a scientific interest group hosted by Inria and including CNRS, RENATER and several Universities as well as other organizations (see https://www.grid5000.fr).
-
All experiments have been performed on Grid’5000 using a Tesla P100 GPU.
-
The codes involving GPs and DGPs are based on Tensorflow (Abadi et al. 2016), GPflow (De Matthews et al. 2017) (https://github.com/GPflow/GPflow), and Doubly-Stochastic-DGP (Salimbeni and Deisenroth 2017) (https://github.com/ICL-SML/Doubly-Stochastic-DGP) in Python 3.
-
The data is always normalized and standardized (zero mean and a variance equal to 1).
-
For all DGPs, ARD RBF (Automatic Relevance Determination - Radial Basis Function) kernels are used with a length-scale and variance initialized to 1 if it does not get an initialization from a previous DGP.
-
The number of Gibbs sampling iterations used is 4 and in each iteration 1000 samples are drawn.
-
LMC is used with a coregionalization matrix of rank 2.
-
The Python package Platypus (Hadka 2015) (https://github.com/Project-Platypus/Platypus) is used for NSGA-II. It is used with its default parameters proposed in Deb et al. (2000) with a population of 5 candidates. The population evolves until reaching \(15\times d\) evaluations of the objective functions, where d is the dimension of the input space.
-
For MO-DGP, the inducing inputs at the different layers are initialized to \({{\textbf {X}}}\).
-
The mean of the variational distribution of the inducing variables for the layer i is initialized at \({{\textbf {y}}}_i\).
-
MO-DGP is optimized in three stages. In the first one, 3000 Adam optimization steps are performed while fixing the variational parameters and the inducing inputs. In the second one, the inducing inputs are also optimized using 3000 Adam optimization steps. Then, 15,000 iterations of the algorithm are performed.
Rights and permissions
About this article
Cite this article
Hebbal, A., Balesdent, M., Brevault, L. et al. Deep Gaussian process for multi-objective Bayesian optimization. Optim Eng 24, 1809–1848 (2023). https://doi.org/10.1007/s11081-022-09753-0
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11081-022-09753-0