Abstract
We consider nonlinear optimization problems that involve surrogate models represented by neural networks. We demonstrate first how to directly embed neural network evaluation into optimization models, highlight a difficulty with this approach that can prevent convergence, and then characterize stationarity of such models. We then present two alternative formulations of these problems in the specific case of feedforward neural networks with ReLU activation: as a mixed-integer optimization problem and as a mathematical program with complementarity constraints. For the latter formulation we prove that stationarity at a point for this problem corresponds to stationarity of the embedded formulation. Each of these formulations may be solved with state-of-the-art optimization methods, and we show how to obtain good initial feasible solutions for these methods. We compare our formulations on three practical applications arising in the design and control of combustion engines, in the generation of adversarial attacks on classifier networks, and in the determination of optimal flows in an oil well network.
Similar content being viewed by others
Data availability
The engine design and adversarial attack generation datasets analyzed in Sections 5.1 and 5.2 are available from the corresponding author upon request. The oil well dataset analyzed in Section 5.3 was provided by the authors of [28] and can be found at the following repository: https://github.com/bgrimstad/relu-opt-public.
References
Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghemawat, S., Irving, G., Isard, M., et al.: TensorFlow: a system for large-scale machine learning. In: 12th \(\{\)USENIX\(\}\) Symposium on Operating Systems Design and Implementation (\(\{\)OSDI\(\}\) 16), 265–283 (2016)
Aithal, SM., Balaprakash, P.: MaLTESE: Large-scale simulation-driven machine learning for transient driving cycles. In: High Performance Computing, 186–205. Springer, Cham (2019)
Anderson, R., Huchette, J., Tjandraatmadja, C., Vielma, JP.: Strong mixed-integer programming formulations for trained neural networks. In: International Conference on Integer Programming and Combinatorial Optimization, 27–42 (2019)
Baumrucker, B., Renfro, J., Biegler, L.: Mpec problem formulations and solution strategies with chemical engineering applications. Comput. Chem. Eng. 32(12), 2903–2913 (2008)
Belotti, P.: Couenne: A user’s manual. Technical report, FICO (2020)
Bergman, D., Huang, T., Brooks, P., Lodi, A., Raghunathan, AU.: Janos: an integrated predictive and prescriptive modeling framework. INFORMS J. Comput. (2021)
Bolte, J., Pauwels, E.: Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning. Math. Program. 1–33 (2020)
Bonami, P., Lee, J.: BONMIN user’s manual. Numer. Math. 4, 1–32 (2007)
Carlini, N., Wagner, D.: Towards evaluating the robustness of neural networks. In: 2017 IEEE Symposium on Security and Privacy (SP), 39–57 (2017)
Cheng, CH., Nührenberg, G., Ruess, H.: Maximum resilience of artificial neural networks. In: International Symposium on Automated Technology for Verification and Analysis, 251–268. Springer, (2017)
Cheon, MS.: An outer-approximation guided optimization approach for constrained neural network inverse problems. Math. Program. 1–30 (2021)
Clarke, L., Linderoth, J., Johnson, E., Nemhauser, G., Bhagavan, R., Jordan, M.: Using OSL to improve the computational results of a MIP logistics model. EKKNEWS 16 (1996)
Delarue, A., Anderson, R., Tjandraatmadja, C.: Reinforcement learning with combinatorial actions: an application to vehicle routing. Adv. Neural Inf. Process. Syst. 33 (2020)
Du, SS., Zhai, X., Poczos, B., Singh, A.: Gradient descent provably optimizes over-parameterized neural networks. In: International Conference on Learning Representations, (2018)
Dunning, I., Huchette, J., Lubin, M.: Jump: a modeling language for mathematical optimization. SIAM Rev. 59(2), 295–320 (2017)
Duran, M., Grossmann, I.: A mixed-integer nonlinear programming algorithm for process systems synthesis. AIChE J. 32(4), 592–606 (1986)
Duran, M.A., Grossmann, I.: An outer-approximation algorithm for a class of mixed-integer nonlinear programs. Math. Program. 36, 307–339 (1986)
Dutta, S., Jha, S., Sankaranarayanan, S., Tiwari, A.: Output range analysis for deep feedforward neural networks. In: NASA Formal Methods Symposium, pp. 121–138. Springer, (2018)
Fischetti, M., Jo, J.: Deep neural networks and mixed integer linear optimization. Constraints 23(3), 296–309 (2018)
Fletcher, R., Leyffer, S.: Solving mathematical program with complementarity constraints as nonlinear programs. Optim. Methods Softw. 19(1), 15–40 (2004)
Fletcher, R., Leyffer, S., Ralph, D., Scholtes, S.: Local convergence of SQP methods for mathematical programs with equilibrium constraints. SIAM J. Optim. 17(1), 259–286 (2006)
Fourer, R., Gay, DM., Kernighan, BW.: AMPL: A Modeling Language for Mathematical Programming. The Scientific Press (1993)
Gale, D.: Neighborly and cyclic polytopes. In: Proc. Sympos. Pure Math 7, pp. 225–232 (1963)
Gleixner, A.M., Berthold, T., Müller, B., Weltge, S.: Three enhancements for optimization-based bound tightening. J. Global Optim. 67(4), 731–757 (2017)
Glorot, X., Bordes, A., Bengio, Y.: Deep sparse rectifier neural networks. In: Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, 315–323. JMLR Workshop and Conference Proceedings, (2011)
Goodfellow, I., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples. In: International Conference on Learning Representations, (2015)
Goodfellow, IJ., Vinyals, O., Saxe, AM.: Qualitatively characterizing neural network optimization problems. arXiv preprintarXiv:1412.6544, (2014)
Grimstad, B., Andersson, H.: ReLU networks as surrogate models in mixed-integer linear programs. Comput. Chem. Eng. 131, 106580 (2019)
Gurobi optimizer reference manual, version 5.0. Gurobi Optim. Inc. (2012)
He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778 (2016)
IBM Corp. IBM Ilog CPLEX V12.1: User’s Manual for CPLEX, (2009)
Katz, G., Barrett, C., Dill, DL., Julian, K., Kochenderfer, MJ.: Reluplex: An efficient SMT solver for verifying deep neural networks. In: International Conference on Computer Aided Verification, pp. 97–117. Springer, (2017)
Khalil, EB., Gupta, A., Dilkina, B.: Combinatorial attacks on binarized neural networks. In: International Conference on Learning Representations, (2018)
Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. Adv. Neural. Inf. Process. Syst. 25, 1097–1105 (2012)
Kurakin, A., Goodfellow, IJ., Bengio, S.: Adversarial examples in the physical world. In: Artificial Intelligence Safety and Security, pp. 99–112. Chapman and Hall/CRC, (2018)
LeCun, Y.: The MNIST database of handwritten digits. http://yann.lecun.com/exdb/mnist/, (1998)
Leyffer, S.: Mathematical programs with complementarity constraints. SIAG/OPT Views News 14(1), 15–18 (2003)
Leyffer, S., Lopez-Calva, G., Nocedal, J.: Interior methods for mathematical programs with complementarity constraints. SIAM J. Optim. 17(1), 52–77 (2006)
Li, Y., Yuan, Y.: Convergence analysis of two-layer neural networks with ReLU activation. Adv. Neural. Inf. Process. Syst. 30, 597–607 (2017)
Lombardi, M., Milano, M., Bartolini, A.: Empirical decision model learning. Artif. Intell. 244, 343–367 (2017)
Mahajan, A., Leyffer, S., Linderoth, J., Luedtke, J., Munson, T.: MINOTAUR: a toolkit for solving mixed-integer nonlinear optimization. wiki-page, (2011). http://wiki.mcs.anl.gov/minotaur
Montufar, G.F., Pascanu, R., Cho, K., Bengio, Y.: On the number of linear regions of deep neural networks. Adv. Neural. Inf. Process. Syst. 27, 2924–2932 (2014)
Papalexopoulos, T., Tjandraatmadja, C., Anderson, R., Vielma, JP., Belanger, D.: Constrained discrete black-box optimization using mixed-integer programming. arXiv preprintarXiv:2110.09569, (2021)
Pascanu, R., Montúfar, G., Bengio, Y.: On the number of response regions of deep feed forward networks with piece-wise linear activations. In: International Conference on Learning Representations, (2014)
Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic differentiation in PyTorch. (2017)
Powell, M.: A method for nonlinear constraints in minimization problems in optimization. In: Fletcher R. (ed.) Optimization. Academic Press, (1969)
Queipo, N.V., Haftka, R.T., Shyy, W., Goel, T., Vaidyanathan, R., Tucker, P.K.: Surrogate-based analysis and optimization. Prog. Aerosp. Sci. 41(1), 1–28 (2005)
Raghunathan, A., Biegler, L.T.: An interior point method for mathematical programs with complementarity constraints (MPCCs). SIAM J. Optim. 15(3), 720–750 (2005)
Ramachandran, P., Zoph, B., Le, QV: Searching for activation functions. arXiv preprintarXiv:1710.05941, (2017)
Ryu, M., Chow, Y., Anderson, R., Tjandraatmadja, C., Boutilier, C.: Caql: Continuous action q-learning. In: International Conference on Learning Representations, (2019)
Sahinidis, N.V.: BARON: a general purpose global optimization software package. J. Global Optim. 8(2), 201–205 (1996)
Scheel, H., Scholtes, S.: Mathematical program with complementarity constraints: Stationarity, optimality and sensitivity. Math. Oper. Res. 25, 1–22 (2000)
Schweidtmann, A.M., Mitsos, A.: Deterministic global optimization with artificial neural networks embedded. J. Optim. Theory Appl. 180(3), 925–948 (2019)
Serra, T., Ramalingam, S.: Empirical bounds on linear regions of deep rectifier networks. In: AAAI, pp. 5628–5635 (2020)
Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. In: International Conference on Learning Representations, (2014)
Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., Fergus, R.: Intriguing properties of neural networks. In: 2nd International Conference on Learning Representations, (2014)
Tawarmalani, M., Sahinidis, N.V.: Convexification and Global Optimization in Continuous and Mixed-Integer Nonlinear Programming: Theory, Algorithms, Software, and Applications. Kluwer Academic Publishers, Boston MA (2002)
Tjeng, V., Xiao, K., Tedrake, R.: Evaluating robustness of neural networks with mixed integer programming. arXiv preprintarXiv:1711.07356, (2017)
Tsay, C., Kronqvist, J., Thebelt, A., Misener, R.: Partition-based formulations for mixed-integer optimization of trained relu neural networks. Adv. Neural Inf. Process. Syst. 34, (2021)
Wächter, A., Biegler, L.T.: On the implementation of a primal-dual interior point filter line search algorithm for large-scale nonlinear programming. Math. Program. 106(1), 25–57 (2006)
Zaslavsky, T.: Facing up to arrangements: Face-count formulas for partitions of space by hyperplanes: Face-count formulas for partitions of space by hyperplanes, vol. 154. American Mathematical Soc., (1975)
Zhang, Z., Brand, M.: Convergent block coordinate descent for training tikhonov regularized deep neural networks, (2017)
Acknowledgements
This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research, under Contract DE-AC02-06CH11357. This work was also supported by the U.S. Department of Energy through grant DE-FG02-05ER25694. The first author was also supported through an NSF-MSGI fellowship.
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.
The submitted manuscript has been created by UChicago Argonne, LLC, Operator of Argonne National Laboratory (“Argonne”). Argonne, a U.S. Department of Energy Office of Science laboratory, is operated under Contract No. DE-AC02-06CH11357. The U.S. Government retains for itself, and others acting on its behalf, a paid-up nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and perform publicly and display publicly, by or on behalf of the Government. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan.http://energy.gov/downloads/doe-public-access-plan.
A additional engine design problem results
A additional engine design problem results
In this appendix, we include the fully tabulated results for selected figures from Section 5.1. In Table 7, we include the data from Fig. 9 which records the average time until each solver found the best solution to its respective formulation. In Table 8, we include the data from Fig. 10 which records the average percentage gap between the optimal solution for each formulation and the best-known solution.
We also include results recording the full experiment solve times for the engine design problem. A bar plot depicting the average time over all 10 runs for each number of hidden layers, and number of timesteps on each formulation is presented in Fig. 13. The results are also tabulated in Table 9. These results are distinct from the results presented in the main section of the paper as they include time after the solver finds its optimal iterate which may be large due to convergence issues. This is especially the case for the embedded formulation for which the solver often reaches the iteration limit after finding a good solution. This happens because Ipopt will jump away from this solution as it cannot prove it to be locally optimal (if the solution is in the neighborhood of a nondifferentiable point) and will continue the solve in a different area. The MIP solves also have a potentially large amount of time after it finds its optimal solution. This extra time is used exploring the remainder of the solution tree to determine that there is no other feasible solution better than the best seen so far. The MPCC times presented here are largely the same as when Ipopt finds its optimal solution in these settings, it usually is able to determine that it is locally optimal and terminate.
We also remark that the full experiment time for the embedded formulations is very similar between the warmstarted and non-warmstarted solves. This similarity can be explained by how, in most cases, the solver reaches the iteration limit for both problems and fails to converge to a point. The warmstarted problem generally finds a good solution faster, but in neither setting is Ipopt able to verify that solutions found are locally optimal.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Yang, D., Balaprakash, P. & Leyffer, S. Modeling design and control problems involving neural network surrogates. Comput Optim Appl 83, 759–800 (2022). https://doi.org/10.1007/s10589-022-00404-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10589-022-00404-9