Pith. sign in

REVIEW 4 major objections 4 minor 48 references

Input convex neural networks as surrogates in mathematical optimisation

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read For an input convex neural network over a box, the epigraph together with the concave envelope built from vertex evaluations is exactly the convex hull of the network's graph, giving the strongest continuous relaxation; direct output…

desk verdict Solid theory, over-sold empirics: the convex-hull relaxation is a genuine contribution, but root termination in Case 1 is solver-dependent and 'default surrogate' needs more evidence. read the letter →

arxiv 2608.09707 v1 pith:TM45QOWR submitted 2026-08-10 math.OC cs.LG

Yu Liu , Jan Kronqvist , Fabricio Oliveira This is my paper · ORCID
classification math.OCcs.LG MSC 90C1190C26
keywords inputconvexneuralnetworkssurrogatemodellingLPrelaxationmixed-integerprogramminghullbranch-and-boundconcaveenvelopeepigraphembedding
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Neural networks used as surrogates in optimisation are normally embedded via mixed-integer programs, with one binary variable per ReLU neuron. This paper argues that when the surrogate is an input convex neural network, a network whose output is convex in its inputs, the embedding can be done with plain linear programming. The key result is that over a box domain, the relaxation obtained by sandwiching the network output between its epigraph (lower bound) and a concave envelope built from the 2^n vertex evaluations is exactly the convex hull of the network's graph, the strongest possible continuous relaxation. A branch-and-bound that branches on the input variables then converges without binary variables, and whenever the epigraph embedding is valid it terminates at the root node. If the argument is right, ICNN surrogates can be embedded in optimisation models without the binary activation variables that make large ReLU networks computationally prohibitive.

What carries the argument

The central object is the concave envelope of the ICNN over a box, constructed explicitly as the value function of an LP whose decision variables are the weights of a convex combination of the box's $2^n$ vertices (Proposition 2). This envelope overestimates the convex ICNN, while the epigraph constraints underestimate it; Theorem 1 proves that the two together reproduce the convex hull of the ICNN's graph. The ICNN-BB algorithm carries this machinery across a branch-and-bound tree, branching on input variables and evaluating the ICNN at new vertices as boxes are bisected.

What would settle it

Solve LP (20) at an interior point of a box for any trained ICNN and check whether the resulting value is indeed the least concave function overestimating the network: if a concave overestimator is found that is strictly below the LP value at that point, Theorem 1 fails. The reported 5_25 wine-blending instance, where ICNN-BB explored roughly 72,000 nodes without returning a feasible solution within the 3600 s limit, is the concrete falsifier of the scalability premise.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1: for a trained ICNN $\hat{f}$ over a box $X$, the set of points $(x,y)$ satisfying $\hat{f}(x) \le y \le \bar z(x)$, where $\bar z$ is the concave envelope obtained by maximising convex combinations of the $2^n$ vertex values of $X$, equals the convex hull of the graph $\{(x,\hat{f}(x)) : x \in X\}$. Together with the exactness of the LP relaxation under direct output minimisation (Corollary 2), this means an ICNN surrogate can be embedded either as an exact LP when the objective penalises overestimation, or as the tightest possible LP relaxation in general, eliminating binary variables in both cases. The construction is tractable specifically because convexity lets the concave envelope be read off from vertex evaluations; no such characterization is known for general ReLU networks.

Load-bearing premise

The practical value of the algorithm rests on the input dimension being small enough that evaluating the network at all $2^n$ vertices of every sub-box remains affordable; the paper's own wine-blending experiment, stacking five 11-dimensional surrogates into 55 dimensions, timed out without returning a feasible solution.

Editorial extensions

If this is right

  • For direct output minimisation of an ICNN, the MIP's binary activation variables can be relaxed to continuous ones, so the surrogate embedding becomes an equivalent LP.
  • Over any box domain, the epigraph plus the vertex-built concave envelope yields the convex hull relaxation of the ICNN's graph, the strongest continuous relaxation available for that surrogate.
  • The branch-and-bound algorithm branches on the input variables rather than on neuron activations, so its branching dimension is the surrogate's input dimension, not the network's size.
  • Whenever the epigraph embedding is valid, the algorithm terminates at the root node without branching, making solve time nearly independent of network width in the experiments.
  • In the food-aid case study, ICNN-MIP and ICNN-BB solved all architectures at the root node, while FNN-MIP timed out on most networks with 50 or more neurons per layer.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An a priori validity check for the epigraph embedding is likely to be easy in many practical problems: if the downstream objective and constraints are non-decreasing in the surrogate output, any slack in $y \ge \hat{f}(x)$ can be removed, so the LP embedding is exact and no branching is needed.
  • The convex-hull construction is not tied to the ICNN architecture per se; it should carry over to any convex piecewise-linear or convex Lipschitz function whose values at the $2^n$ vertices are available, which suggests immediate extensions to other convex surrogates such as convex quadratic models or kernel-based convex regressors.
  • The paper's open direction of sparsified or hierarchical vertex sets translates naturally into a testable trade-off: sampling fewer than $2^n$ vertices yields cheaper but weaker relaxations, and the empirical crossover point at roughly eleven input dimensions in the wine-blending study gives a concrete benchmark for when such approximations would be needed.
  • The 5_25 wine-blending timeout, with 55 stacked dimensions, indicates that a stacked deployment of many low-dimensional surrogates is more practical than a single high-dimensional surrogate, because vertex enumeration cost is per surrogate rather than over the product domain.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper studies embedding trained input convex neural networks (ICNNs) as surrogates in mathematical optimisation problems. It proves that the LP relaxation of the ICNN-MIP formulation recovers the exact ICNN output for fixed inputs and under direct output minimisation (Proposition 1, Corollaries 1 and 2). It formalises when the epigraph embedding is a valid reformulation (Definition 1), constructs a concave envelope from vertex evaluations (Proposition 2), and proves that adding this envelope to the epigraph yields the convex hull of the ICNN's graph over a box (Theorem 1). On this basis it proposes ICNN-BB, a branch-and-bound algorithm that branches on the surrogate input variables rather than on neural-network activation binaries, and reports three case studies: humanitarian food aid, oil well routing, and wine blending. The theoretical core is largely correct, but the empirical support for several headline practical claims is incomplete, and the algorithm's specification is ambiguous for problem instances with outer integer variables.

Significance. If the results stand, the paper makes a substantial contribution: it provides a tractable construction of the convex hull of the graph of an entire ReLU-based network under input convexity, which is not known for general ReLU networks, and it demonstrates a practical way to avoid binary activation variables for ICNN surrogates. The proofs of Proposition 1, Proposition 2, and Theorem 1 are self-contained and check out, and the manuscript ships reproducibility materials. However, the practical claims of root-node termination and of an LP-only search tree are not fully supported by the theory as written, and the reported experiments therefore need careful re-interpretation before the paper can be accepted.

major comments (4)
  1. [Section 4.3, Definition 1, and Case 1 (Eq. (32), Tables 2-3)] The root-node termination reported in Tables 2 and 3 is not established by the theory. In the food-aid model the objective (32a) is independent of y and the only y-constraint is the lower bound y >= 0.5 in (32h), so in the root epigraph relaxation (19) y has no upper bound and the LP has many optimal solutions with y > fhat(x). Thus Definition 1(ii) is not satisfied, and the gap check in Algorithm 1 (line 13) is solver-dependent: the algorithm terminates at the root only if the LP solver happens to return an optimal solution with y equal to the ICNN forward-pass value. The paper should either prove that the particular solvers used return such a basic solution for these degenerate LPs, or weaken the claim that ICNN-BB terminates at the root whenever the epigraph embedding is valid, and re-interpret the Case 1 tables accordingly.
  2. [Section 4.3, Algorithm 1, and Case 2 (Eq. (33), Tables 5-6)] Algorithm 1 is written for the continuous problem (4), but the oil-well routing problem in Case 2 contains binary routing variables y_e in (33l). The paper does not specify how ICNN-BB treats these outer integer variables. If each node subproblem is solved as a mixed-integer program, then the statement in Section 4.3 that the algorithm 'achieves convergence via LP solves throughout the search tree' is inaccurate, and the reported ICNN-BB node counts and per-node times are not directly comparable with those of the MIP baselines unless the inner branching is also reported. The authors should clarify whether outer binaries are handled by an inner MIP solve, by a unified branch-and-bound tree, or by further relaxation, and should report node counts accordingly.
  3. [Section 4.2.2, Theorem 1, and Section 4.3 (stacked surrogates)] Theorem 1 concerns a single ICNN over its box. In the stacked-surrogate setting used in Case 3, the per-surrogate envelope blocks use independent alpha variables for each surrogate, so the resulting feasible set is the product of per-surrogate hulls rather than the convex hull of the stacked graph. The text should state this explicitly; as written, a reader could infer that Theorem 1 justifies the stacked relaxation as the strongest continuous relaxation of the full multi-surrogate problem.
  4. [Section 4.3, Algorithm 1 (termination criterion), and Section 5.2-5.4] The algorithm's acceptance test in line 14 accepts any relaxation solution with relative gap epsilon <= tau, so the returned point is only approximately feasible for the original constraint y = fhat(x), not exactly feasible. This matters for the reported objectives: in Case 2, for example, the surrogate output enters the objective through (33h), so a one-percent overestimate can inflate the reported oil flow. The paper should state that ICNN-BB returns an approximately feasible solution by design, and should discuss the implications for the numerical comparisons with the exact MIP formulations.
minor comments (4)
  1. [Table 1] The ICNN MSE for the 3_50 network is listed as 9.652e-4, identical to the 2_50 ICNN MSE; if this is not a typo, the duplicate value should be corrected.
  2. [Section 5.4.3 (5_25 row)] The statement that 'the vertex set and LP size grow rapidly with branching depth' is inaccurate: a sub-box of an 11-dimensional box still has 2^11 vertices as long as no coordinate is fixed, so the per-node LP size does not grow with depth. What grows is the number of nodes and hence the total number of vertex evaluations.
  3. [Abstract and Section 6] The abstract's closing claim that the results support ICNN 'as the default surrogate' is stronger than the experiments warrant, since on the 5_25 wine-blending instance ICNN-BB fails to return a feasible solution and ICNN-MIP is slower than FNN-MIP under Gurobi; the more careful qualification in Section 6 about the input-dimensionality frontier should be reflected in the abstract.
  4. [Section 3.3] The illustrative comparison of FNN and ICNN surrogates for f(x)=x^2 states that the two networks have 'comparable accuracy' but does not report the training or validation errors; adding these numbers would strengthen the point.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the convex-hull relaxation, LP exactness results, and ICNN-BB convergence are derived from convexity and the network equations, not from fitted quantities or load-bearing self-citations.

full rationale

The paper's central derivations are self-contained. Proposition 1 and Corollaries 1–2 prove LP exactness for fixed and optimized inputs by induction over layers using nonnegative weights, so no fitted parameter is renamed as a prediction. Proposition 2 constructs the concave envelope from 2^n vertex forward passes, and Theorem 1 proves the projected feasible set equals the convex hull of the ICNN graph by sandwiching the epigraph below and the concave envelope above; both arguments use only the definition of convexity and standard polyhedral facts. The vertex values used in the envelope are outputs of the trained ICNN, not constants fitted to the downstream objective, so the relaxation is not circular. The statement that the root node terminates when the epigraph embedding is valid follows from Definition 1(ii) together with Algorithm 1's forward-pass gap check; this is a designed algorithmic consequence rather than an independent empirical prediction, and it is not used to derive any numerical result. The Case 1 root termination is reported as an observed solver outcome, and the paper also candidly reports the scalability limitation in Case 3, where ICNN-BB fails to return a feasible solution on the 55-dimensional instance. Self-citations to Liu et al. (2025) and Liu and Oliveira (2025) are contextual and not load-bearing; the ICNN convexity and LP-inference properties are cited to Amos et al. (2017), an external source. A possible degeneracy of the root LP when y is absent from the objective is a correctness or robustness concern, not a circularity concern. Therefore no circular step is present.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The theoretical claims rest on standard convex analysis and LP/MIP theory plus the ICNN architecture's convexity guarantee. The practical claims additionally depend on approximation quality of the trained ICNN, small input dimension for vertex enumeration, and an a posteriori check of epigraph validity. No free parameter is fitted to make the theorems true.

free parameters (3)
  • branch-and-bound optimality tolerance tau = 0.01
    Chosen across all case studies in Section 5.1; controls when a relaxed solution is accepted as optimal.
  • training hyperparameters (learning rates, epochs, batch sizes) = varies per case study, e.g., ICNN learning rate 0.0015 in Case 1, 0.001 in Cases 2 and 3
    Hand-selected per case study; they affect approximation quality but not the theoretical derivations.
  • network architectures (layers and neurons) = 2_10 to 3_100 for Case 1; 2x20 for wells and 2x50 for riser in Case 2; 3x20 in Case 3
    Chosen to match FNN and ICNN capacity; not fitted to the optimization claim.
assumptions (5)
  • domain assumption ICNNs with nonnegative weights and convex nondecreasing activations produce convex functions (Amos et al. 2017).
    Used throughout Section 2.2 and in Proposition 1 and Theorem 1 to justify the induction and convexity arguments.
  • domain assumption The learned surrogate approximately matches the true underlying response, with small approximation error.
    Needed for the practical claim that ICNN surrogates are accurate; Case 2 riser shows an order of magnitude larger MSE than FNN.
  • domain assumption The concave envelope converges uniformly to the ICNN as sub-boxes shrink.
    Asserted in Section 4.3 for finite termination of ICNN-BB; not proven in the paper, though true for continuous piecewise-linear functions.
  • domain assumption The input domain is a box and activations are ReLU.
    Used in Proposition 2 and Theorem 1; vertex enumeration requires a box.
  • standard math Standard results of linear programming and mixed-integer programming, including big-M formulations and LP relaxations.
    Implicit in the MIP formulations and solver usage throughout the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Input convex neural networks as surrogates in mathematical optimisation." pith.science (2026). https://pith.science/paper/TM45QOWR

@misc{pith2026260809707,
  author       = {Pith},
  title        = {Pith review of: Input convex neural networks as surrogates in mathematical optimisation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TM45QOWR}},
  note         = {Machine review of arXiv:2608.09707}
}
read the original abstract

Embedding trained neural networks as surrogates within optimisation problems is an established practice in operations research. The prevailing approach uses feedforward neural networks (FNNs) with ReLU activations, whose piecewise-linear structure admits an exact but computationally intensive mixed-integer programming (MIP) reformulation as the networks grow. We advocate input convex neural networks (ICNNs) as structurally superior surrogates when the underlying response is approximately convex or concave. The convex architecture offers two computational advantages. First, the ICNN-MIP formulation tends to yield a tighter linear programming (LP) relaxation than its FNN-MIP counterpart, with no integrality gap in favourable instances. Second, ICNNs uniquely admit an LP-based reformulation via epigraph representations of ReLU activations, though this embedding is not always exact. When it is not, we exploit the properties of ICNNs to construct the strongest continuous relaxation over box domains, namely, the convex hull of the ICNN's graph, bounded below by the epigraph and above by the concave envelope; this construction is tractable under input convexity but hard for general ReLU networks. On this basis, we develop a branch-and-bound algorithm that builds this relaxation at each node, branches directly on input variables rather than intermediate variables as in MIP reformulations, and terminates at the root node whenever the epigraph embedding is valid. Case studies on humanitarian food aid, oil well routing, and wine blending show that ICNN surrogates match FNN accuracy and deliver gains in solve time and scalability, supporting ICNN as the default surrogate when the underlying function is convex, concave, or well-approximated as such.

Figures

Figures reproduced from arXiv: 2608.09707 by the authors.

Figure 1
Figure 1. A simplified illustration of the ICNN architecture, reproduced from [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Neural surrogates and their LP relaxations for [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Network structure of the oil production problem with two manifolds and eight wells, adapted from [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 18 canonical work pages

  1. [1]

    Addis, B., Castel, C., Macali, A., Misener, R., & Piccialli, V. (2023). Data augmentation driven by optimization for membrane separation process synthesis. Computers & Chemical Engineering , 177, 108342. https://doi.org/10.1016/j.compchemeng.2023.108342

    arXiv 2023

  2. [2]

    Amos, B., Xu, L., & Kolter, J. Z. (2017). Input Convex Neural Networks . https://doi.org/10.48550/arXiv.1609.07152

    arXiv doi:10.48550/arxiv.1609.07152 2017

  3. [3]

    Anderson, R., Huchette, J., Ma, W., Tjandraatmadja, C., & Vielma, J. P. (2020). Strong mixed-integer programming formulations for trained neural networks. Mathematical Programming , 183(1), 3--39. https://doi.org/10.1007/s10107-020-01474-5

    doi:10.1007/s10107-020-01474-5 2020

  4. [4]

    Bezanson, J., Edelman, A., Karpinski, S., & Shah, V. B. (2017). Julia: A Fresh Approach to Numerical Computing . SIAM Review , 59(1), 65--98. https://doi.org/10.1137/141000671

    doi:10.1137/141000671 2017

  5. [5]

    Bünning, F., Schalbetter, A., Aboudonia, A., Badyn, M. H. d., Heer, P., & Lygeros, J. (2021). Input Convex Neural Networks for Building MPC . Proceedings of the 3rd Conference on Learning for Dynamics and Control , 251--262. https://proceedings.mlr.press/v144/bunning21a.html

    work page 2021

  6. [6]

    Chen, Y., Shi, Y., & Zhang, B. (2020). Data- Driven Optimal Voltage Regulation Using Input Convex Neural Networks . Electric Power Systems Research , 189, 106741. https://doi.org/10.1016/j.epsr.2020.106741

    arXiv 2020

  7. [7]

    Cheng, C.-H., Nührenberg, G., & Ruess, H. (2017). Maximum Resilience of Artificial Neural Networks . https://doi.org/10.48550/arXiv.1705.01040

    work page Pith review arXiv doi:10.48550/arxiv.1705.01040 2017

  8. [8]

    Cheng, Y., Wang, D., Zhou, P., & Zhang, T. (2020). A Survey of Model Compression and Acceleration for Deep Neural Networks . https://doi.org/10.48550/arXiv.1710.09282

    arXiv doi:10.48550/arxiv.1710.09282 2020

Show all 48 references
  1. [9]

    Cortez, P., Cerdeira, A., Almeida, F., Matos, T., & Reis, J. (2009). Modeling wine preferences by data mining from physicochemical properties. Decision Support Systems , 47(4), 547--553. https://doi.org/10.1016/j.dss.2009.05.016

    2009 doi

  2. [10]

    B., & Bent, R

    Dowson, O., Parker, R. B., & Bent, R. (2026). MathOptAI .jl: Embed Trained Machine - Learning Predictors into JuMP Models . INFORMS Journal on Computing . https://doi.org/10.1287/ijoc.2025.1446

    2026

  3. [11]

    Dvorkin, V., Chevalier, S., & Chatzivasileiadis, S. (2023). Emission- Constrained Optimization of Gas Networks : Input - Convex Neural Network Approach . 2023 62nd IEEE Conference on Decision and Control ( CDC ) , 1575--1579. https://doi.org/10.1109/CDC49753.2023.10383948

    2023

  4. [12]

    Fischetti, M. & Jo, J. (2018). Deep neural networks and mixed integer linear optimization. Constraints , 23(3), 296--309. https://doi.org/10.1007/s10601-018-9285-6

    2018 doi

  5. [13]

    Fons Moreno-Palancas, I., Ruiz Femenia, R., Salcedo Díaz, R., & Caballero, J. A. (2025). Solving bilevel problems under uncertainty with embedded neural networks: Incorporating scenario sets as inputs. Digital Chemical Engineering , 16, 100253. https://doi.org/10.1016/j.dche.2...

    2025

  6. [14]

    Gagneux, A., Massias, M., Soubies, E., & Gribonval, R. (2025). Convexity in ReLU Neural Networks : Beyond ICNNs ? Journal of Mathematical Imaging and Vision , 67(4), 40. https://doi.org/10.1007/s10851-025-01253-x

    2025 doi

  7. [15]

    Gokhale, G., Claessens, B., & Develder, C. (2022). Physics informed neural networks for control oriented thermal modeling of buildings. Applied Energy , 314, 118852. https://doi.org/10.1016/j.apenergy.2022.118852

    2022

  8. [16]

    Grimstad, B. (2022). bgrimstad/relu-opt-public: Code for optimizing milp problems with embedded relu networks . https://github.com/bgrimstad/relu-opt-public

    2022

  9. [17]

    & Andersson, H

    Grimstad, B. & Andersson, H. (2019). ReLU networks as surrogate models in mixed-integer linear programs. Computers & Chemical Engineering , 131, 106580. https://doi.org/10.1016/j.compchemeng.2019.106580

    2019

  10. [18]

    Gurobi Optimizer Reference Manual

    Gurobi Optimization, LLC (2024). Gurobi Optimizer Reference Manual . https://www.gurobi.com

    2024

  11. [19]

    Hornik, K., Stinchcombe, M., & White, H. (1989). Multilayer feedforward networks are universal approximators. Neural Networks , 2(5), 359--366. https://doi.org/10.1016/0893-6080(89)90020-8

    1989 doi

  12. [20]

    & Hall, J

    Huangfu, Q. & Hall, J. A. J. (2018). Parallelizing the dual revised simplex method. Mathematical Programming Computation , 10(1), 119--142. https://doi.org/10.1007/s12532-017-0130-5

    2018 doi

  13. [21]

    Huchette, J., Muñoz, G., Serra, T., & Tsay, C. (2026). When Deep Learning Meets Polyhedral Theory : A Survey . INFORMS Journal on Computing , 0(0). https://doi.org/10.1287/ijoc.2024.0902

    2026

  14. [22]

    Innes, M. (2018). Flux: Elegant machine learning with Julia . Journal of Open Source Software , 3(25), 602. https://doi.org/10.21105/joss.00602

    2018 doi

  15. [23]

    Jiang, K., Hu, C., & Yan, F. (2022). Path-following control of autonomous ground vehicles based on input convex neural networks. Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering , 236(13), 2806--2816. https://doi.org/10.1177/095...

    2022 doi

  16. [24]

    S., Ali, M., & Pistikopoulos, E

    Kenefake, D., Kakodkar, R., Akundi, S. S., Ali, M., & Pistikopoulos, E. N. (2024). A multiparametric approach to accelerating ReLU neural network based model predictive control. Control Engineering Practice , 151, 106041. https://doi.org/10.1016/j.conengprac.2024.106041

    2024

  17. [25]

    H., Kwon, H., & Jeong, H

    Kim, D., Jang, S., Lee, J.-i., Rhee, S. H., Kwon, H., & Jeong, H. (2026). Ship operational data driven fuel efficiency assessment for shaft generator using input convex neural network. Energy Reports , 15, 108993. https://doi.org/10.1016/j.egyr.2025.108993

    2026

  18. [26]

    Kingma, D. P. & Ba, J. (2017). Adam: A Method for Stochastic Optimization . https://doi.org/10.48550/arXiv.1412.6980

    2017 doi arXiv

  19. [27]

    Kronqvist, J., Misener, R., & Tsay, C. (2025). P-split formulations: a class of intermediate formulations between big- M and convex hull for disjunctive constraints. Mathematical Programming . https://doi.org/10.1007/s10107-025-02232-1

    2025 doi

  20. [28]

    & Oliveira, F

    Liu, Y. & Oliveira, F. (2025). Simulator-based surrogate optimisation employing adaptive uncertainty-aware sampling. Computers & Chemical Engineering , 201, 109243. https://doi.org/10.1016/j.compchemeng.2025.109243

    2025

  21. [29]

    Liu, Y., Oliveira, F., & Kronqvist, J. (2025). ICNN -enhanced 2SP : Leveraging input convex neural networks for solving two-stage stochastic programming . https://doi.org/10.48550/arXiv.2505.05261

    2025 doi arXiv

  22. [30]

    Lubin, M., Dowson, O., Dias Garcia , J., Huchette, J., Legat, B., & Vielma, J. P. (2023). JuMP 1.0: R ecent improvements to a modeling language for mathematical optimization. Mathematical Programming Computation . https://doi.org/10.1007/s12532-023-00239-3

    2023 doi

  23. [31]

    V., Taghvaei, A., Oh, S., & Lee, J

    Makkuva, A. V., Taghvaei, A., Oh, S., & Lee, J. D. (2020). Optimal transport mapping via input convex neural networks . https://doi.org/10.48550/arXiv.1908.10962

    2020 doi arXiv

  24. [32]

    & Wiberg, H

    Maragno, D. & Wiberg, H. (2021). Opticl: Mixed-integer optimization with constraint learning . https://github.com/hwiberg/OptiCL/

    2021

  25. [33]

    \.I ., den Hertog, D., & Fajemisin, A

    Maragno, D., Wiberg, H., Bertsimas, D., Birbil, S . \.I ., den Hertog, D., & Fajemisin, A. O. (2023). Mixed- Integer Optimization with Constraint Learning . Operations Research . https://doi.org/10.1287/opre.2021.0707

    2023

  26. [34]

    Peters, K., Silva, S., Gonçalves, R., Kavelj, M., Fleuren, H., den Hertog, D., Ergun, O., & Freeman, M. (2021). The Nutritious Supply Chain : Optimizing Humanitarian Food Assistance . INFORMS Journal on Optimization , 3(2), 200--226. https://doi.org/10.1287/ijoo.2019.0047

    2021

  27. [35]

    Rosemberg, A., Tanneau, M., Fanzeres, B., Garcia, J., & Van Hentenryck, P. (2024). Learning Optimal Power Flow value functions with input-convex neural networks. Electric Power Systems Research , 235, 110643. https://doi.org/10.1016/j.epsr.2024.110643

    2024

  28. [36]

    Serra, T., Tjandraatmadja, C., & Ramalingam, S. (2018). Bounding and Counting Linear Regions of Deep Neural Networks . https://doi.org/10.48550/arXiv.1711.02114

    2018 doi arXiv

  29. [37]

    Tjeng, V., Xiao, K., & Tedrake, R. (2019). Evaluating Robustness of Neural Networks with Mixed Integer Programming . https://doi.org/10.48550/arXiv.1711.07356

    2019 doi arXiv

  30. [38]

    Tong, J., Cai, J., & Serra, T. (2024). Optimization over Trained Neural Networks : Taking a Relaxing Walk . Integration of Constraint Programming , Artificial Intelligence , and Operations Research , 221--233. https://doi.org/10.1007/978-3-031-60599-4_14

    2024 doi

  31. [39]

    Tsay, C. (2021). Sobolev trained neural network surrogate models for optimization. Computers & Chemical Engineering , 153, 107419. https://doi.org/10.1016/j.compchemeng.2021.107419

    2021

  32. [40]

    Turner, M., Chmiela, A., Koch, T., & Winkler, M. (2024). SurrogateLIB : An extendable library of mixed-integer programs with embedded machine learning predictors . https://zenodo.org/records/11231147

    2024

  33. [41]

    Wang, C., Ma, S., Zha, B., & Ma, B. (2025). Surrogate- Accelerated Evolutionary Algorithm for Solving Electric Vehicle Routing Problems . IEEE Transactions on Transportation Electrification , 11(1), 3524--3537. https://doi.org/10.1109/TTE.2024.3443515

    2025

  34. [42]

    Wang, K., Lozano, L., Cardonha, C., & Bergman, D. (2023). Optimizing over an Ensemble of Trained Neural Networks . INFORMS Journal on Computing , 35(3), 652--674. https://doi.org/10.1287/ijoc.2023.1285

    2023

  35. [43]

    Wang, W., Zhang, H., Wang, Y., Tian, Y., & Wu, Z. (2024). Fast Explicit Machine Learning - Based Model Predictive Control of Nonlinear Processes Using Input Convex Neural Networks . Industrial & Engineering Chemistry Research , 63(40), 17279--17293. https://doi.org/10.1021/acs...

    2024 doi

  36. [44]

    & Wang, J

    Wu, T. & Wang, J. (2024). Transient stability-constrained unit commitment using input convex neural network. IEEE Transactions on Neural Networks and Learning Systems , 35(11), 16023--16035. https://doi.org/10.1109/TNNLS.2023.3291673

    2024

  37. [45]

    & Bequette, B

    Yang, S. & Bequette, B. W. (2021). Optimization-based control using input convex neural networks. Computers & Chemical Engineering , 144, 107143. https://doi.org/10.1016/j.compchemeng.2020.107143

    2021

  38. [46]

    Zheng, Y., Hu, C., Wang, X., & Wu, Z. (2023). Physics-informed recurrent neural network modeling for predictive control of nonlinear processes. Journal of Process Control , 128, 103005. https://doi.org/10.1016/j.jprocont.2023.103005

    2023

  39. [47]

    Zhu, J., Xu, Y., Tai, N., Xie, Y., Wen, Q., & Sun, H. (2025). Day-ahead joint chance-constrained optimal scheduling for distribution networks using partial input convex neural networks. IEEE Transactions on Smart Grid , 1--1. https://doi.org/10.1109/TSG.2025.3616160

    2025

  40. [48]

    Ławryńczuk, M. (2022). Input convex neural networks in nonlinear predictive control: A multi-model approach. Neurocomputing , 513, 273--293. https://doi.org/10.1016/j.neucom.2022.09.108

    2022 doi

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.