Abstract
For a general production network, state-of-the-art methods for constructing sparse flexible designs are heuristic in nature, typically computing a proxy for the quality of unseen networks and using that estimate in a greedy manner to modify a current design. This paper develops two machine learning-based approaches to constructing sparse flexible designs that leverage a neural network to accurately and quickly predict the performance of large numbers of candidate designs. We demonstrate that our heuristics are competitive with existing approaches and produce high-quality solutions for both balanced and unbalanced networks. Finally, we introduce a novel application of process flexibility in healthcare operations to demonstrate the effectiveness of our approach in a large numerical case study. We study the flexibility of linear accelerators that deliver radiation to treat various types of cancer. We demonstrate how clinical constraints can be easily absorbed into the machine learning subroutine and how our sparse flexible treatment networks meet or beat the performance of those designed by state-of-the-art methods.
Similar content being viewed by others
References
Akşin OZ, Karaesmen F (2007) Characterizing the performance of process flexibility structures. Oper Res Lett 35(4):477–484
Applegate DL, Bixby RE, Chvátal V, Cook W, Espinoza DG, Goycoolea M, Helsgaun K (2009) Certification of an optimal tsp tour through 85,900 cities. Oper Res Lett 37(1):11–15
Atun R, Jaffray DA, Barton MB, Bray F, Baumann M, Vikram B, Hanna TP, Knaul FM, Lievens Y, Lui TYM et al (2015) Expanding global access to radiotherapy. Lancet Oncol 16(10):1153–1186
Bassamboo A, Randhawa RS, Mieghem JAV (2012) A little flexibility is all you need: on the asymptotic value of flexible capacity in parallel queuing systems. Oper Res 60(6):1423–1435
Bello I, Pham H, Le QV, Norouzi M, Bengio S (2016) Neural combinatorial optimization with reinforcement learning. arXiv preprint arXiv:161109940
Bengio Y, Lodi A, Prouvost A (2018) Machine learning for combinatorial optimization: a methodological tour d’horizon. arXiv preprint arXiv:181106128
Bikker IA, Kortbeek N, van Os RM, Boucherie RJ (2015) Reducing access times for radiation treatment by aligning the doctor’s schemes. Oper Res Health Care 7:111–121
Cavalcante IM, Frazzon EM, Forcellini FA, Ivanov D (2019) A supervised machine learning approach to data-driven simulation of resilient supplier selection in digital manufacturing. Int J Inf Manage 49:86–97
Chan CW, Huang M, Sarhangian V (2021a) Dynamic server assignment in multiclass queues with shifts, with applications to nurse staffing in emergency departments. Oper Res
Chan CW, Sarhangian V, Talwai P, Gogia K (2021b) Utilizing partial flexibility to improve emergency department flow: theory and implementation. Working Paper http://www.columbia.edu/~cc3179/EDStaffingTrial_2020.pdf
Chan TCY, Fearing D (2019) Process flexibility in baseball: the value of positional flexibility. Manage Sci 65(4):1642–1666
Chen S, Song JSJ, Wei Y (2020) Data-driven scalable e-commerce transportation network design with unknown flow response. Available at SSRN 3590865
Chen X, Ma T, Zhang J, Zhou Y (2019) Optimal design of process flexibility for general production systems. Oper Res 67(2):516–531
Chou MC, Teo CP, Zheng H (2008) Process flexibility: design, evaluation, and applications. Flex Serv Manuf J 20(1–2):59–94
Chou MC, Chua GA, Teo CP (2010) On range and response: dimensions of process flexibility. Eur J Oper Res 207(2):711–724
Chou MC, Chua GA, Teo CP, Zheng H (2010) Design for process flexibility: efficiency of the long chain and sparse structure. Oper Res 58(1):43–58
Chou MC, Chua GA, Teo CP, Zheng H (2011) Process flexibility revisited: the graph expander and its applications. Oper Res 59(5):1090–1105
Delaney G, Jacob S, Featherstone C, Barton M (2005) The role of radiotherapy in cancer treatment: Estimating optimal utilization from a review of evidence-based clinical guidelines. Cancer: Interdiscip Int J Am Cancer Soc 104(6):1129–1137
Deng T, Shen ZJM (2013) Process flexibility design in unbalanced networks. Manuf Serv Oper Manag 15(1):24–32
Désir A, Goyal V, Wei Y, Zhang J (2016) Sparse process flexibility designs: Is the long chain really optimal? Oper Res 64(2):416–431
Dong J, Shi P, Zheng F, Jin X (2019) Off-service placement in inpatient ward network: Resource pooling versus service slowdown. Columbia Business School Research Paper (Forthcoming)
Feng W, Wang C, Shen ZJM (2017) Process flexibility design in heterogeneous and unbalanced networks: a stochastic programming approach. IISE Trans 49(8):781–799
Fischetti M, Fraccaro M (2017) Using OR + AI to predict the optimal production of offshore wind parks: A preliminary study. In: Sterle C (ed) Sforza A. Optimization and Decision Science, Methodologies and Applications, pp 203–211
Graves SC, Tomlin BT (2003) Process flexibility in supply chains. Manage Sci 49(7):907–919
Gupta P, Gasse M, Khalil EB, Kumar MP, Lodi A, Bengio Y (2020) Hybrid models for learning to branch. arXiv preprint arXiv:200615212
Gurumurthi S, Benjaafar S (2004) Modeling and analysis of flexible queueing systems. Naval Res Logist (NRL) 51(5):755–782
Hinton GE, Deng L, Yu D, Dahl GE, Ar Mohamed, Jaitly N, Senior A, Vanhoucke V, Nguyen P, Sainath TN et al (2012) Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups. IEEE Signal Process Mag 29(6):82–97
Hopfield JJ, Tank DW (1985) Neural computation of decisions in optimization problems. Biol Cybern 52(3):141–152
Hopp WJ, Tekin E, Van Oyen MP (2004) Benefits of skill chaining in serial production lines with cross-trained workers. Manage Sci 50(1):83–98
Iravani SM, Van Oyen MP, Sims KT (2005) Structural flexibility: a new perspective on the design of manufacturing and service operations. Manage Sci 51(2):151–166
Jordan WC, Graves SC (1995) Principles on the benefits of manufacturing process flexibility. Manage Sci 41(4):577–594
Joustra PE, Van der Sluis E, Van Dijk NM (2010) To pool or not to pool in hospitals: a theoretical and practical comparison for a radiotherapy outpatient department. Ann Oper Res 178(1):77–89
Joustra PE, Kolfin R, van Dijk NM, Koning CCE, Bakker PJM (2012) Reduce fluctuations in capacity to improve the accessibility of radiotherapy treatment cost-effectively. Flex Serv Manuf J 24(4):448–464
Kaempfer Y, Wolf L (2018) Learning the multiple traveling salesmen problem with permutation invariant pooling networks. arXiv preprint arXiv:180309621
Khalil E, Dai H, Zhang Y, Dilkina B, Song L (2017) Learning combinatorial optimization algorithms over graphs. In: Advances in neural information processing systems, pp 6348–6358
Kingma DP, Ba J (2014) Adam: A method for stochastic optimization. arXiv preprint arXiv:14126980
Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems, pp 1097–1105
Larsen E, Lachapelle S, Bengio Y, Frejinger E, Lacoste-Julien S, Lodi A (2018) Predicting solution summaries to integer linear programs under imperfect information with machine learning. arXiv preprint arXiv:180711876
LeCun Y, Bengio Y, Hinton GE (2015) Deep learning. Nature 521(7553):436
Legrain A, Fortin MA, Lahrichi N, Rousseau LM (2015) Online stochastic optimization of radiotherapy patient scheduling. Health Care Manag Sci 18(2):110–123
Li S, Geng N, Xie X (2015) Radiation queue: meeting patient waiting time targets. IEEE Robot Autom Mag 22(2):51–63
Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, van der Laak JAWM, Van Ginneken B, Sánchez CI (2017) A survey on deep learning in medical image analysis. Med Image Anal 42:60–88
Mak HY, Shen ZJM (2009) Stochastic programming approach to process flexibility design. Flex Serv Manuf J 21(3–4):75–91
Price S, Golden B, Wasil E, Zhang HH (2013) Optimizing throughput of a multi-room proton therapy treatment center via simulation. In: Simulation conference (WSC), 2013 Winter, pp 2422–2431
Priore P, Gomez A, Pino R, Rosillo R (2014) Dynamic scheduling of manufacturing systems using machine learning: an updated review. Ai Edam 28(1):83–97
Priore P, Ponte B, Puente J, Gómez A (2018) Learning-based scheduling of flexible manufacturing systems using ensemble methods. Comput Indu Eng 126:282–291
Saure A, Patrick J, Tyldesley S, Puterman ML (2012) Dynamic multi-appointment patient scheduling for radiation therapy. Eur J Oper Res 223(2):573–584
Seward C (2017) Deep learning for warehouse operations. https://twimlai.com/twiml-talk-038-calvin-seward-deep-learning-warehouse-operations/
Sheikhzadeh M, Benjaafar S, Gupta D (1998) Machine sharing in manufacturing systems: total flexibility versus chaining. Int J Flex Manuf Syst 10(4):351–378
Simchi-Levi D (2010) Operations rules: Delivering customer value through flexible operations. MIT Press
Simchi-Levi D, Wei Y (2012) Understanding the performance of the long chain and sparse designs in process flexibility. Oper Res 60(5):1125–1141
Simchi-Levi D, Wei Y (2015) Worst-case analysis of process flexibility designs. Oper Res 63(1):166–185
Smith KA (1999) Neural networks for combinatorial optimization: a review of more than a decade of research. INFORMS J Comput 11(1):15–34
Song H, Tucker AL, Graue R, Moravick S, Yang JJ (2020) Capacity pooling in hospitals: the hidden consequences of off-service placement. Manage Sci 66(9):3825–3842
Sutskever I, Vinyals O, Le QV (2014) Sequence to sequence learning with neural networks. In: Advances in neural information processing systems, pp 3104–3112
Tsitsiklis JN, Xu K (2012) On the power of (even a little) resource pooling. Stochastic Syst 2(1):1–66
Vieira B, Hans EW, van Vliet-Vroegindeweij C, van de Kamer J, van Harten W (2016) Operations research for resource planning and-use in radiotherapy: A literature review. BMC Med Inform Decis Mak 16(1):149
Vinyals O, Fortunato M, Jaitly N (2015) Pointer networks. In: Advances in neural information processing systems, pp 2692–2700
Wallace RB, Whitt W (2005) A staffing algorithm for call centers with skill-based routing. Manuf Serv Oper Manag 7(4):276–294
Wang X, Zhang J (2015) Process flexibility: a distribution-free bound on the performance of k-chain. Oper Res 63(3):555–571
Werker G, Sauré A, French J, Shechter S (2009) The use of discrete-event simulation modelling to improve radiation therapy planning processes. Radiother Oncol 92(1):76–82
Yan Z, Gao SY, Teo CP (2018) On the design of sparse but efficient structures in operations. Manage Sci 64:3421–3445
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.
Appendices
Appendix A: Neural network implementation
1.1 Implementation
We initially considered a convolutional neural network (CNN) with multiple different kernel sizes. We found that when the kernel size equalled the input dimensions, the model had the best performance, thus rendering our CNN implementation equivalent to a fully connected NN. This result is not surprising given the lack of feature locality in our input data.
Our implemented neural network has two hidden layers. The first hidden layer has 1024 \(m \times n\) filters. Each filter is an \(m \times n\) matrix of weights that is multiplied element-wise with the incidence matrix. We found that initializing the filters to be orthogonal matrices with no bias terms improved our model accuracy. The output of each filter is a single value that is stored in a vector of length \(1024 \times 1\). The second hidden layer is a fully connected layer of 128 hidden units, where each unit is connected to every element in the convolutional layer output vector. Dozens of variations of this architecture were evaluated using the validation set before settling on this specification. We used ReLU activation functions.
There are a number of sophisticated open source software packages that support building, training, and deploying neural networks. In this paper, we used the Keras Python package as a wrapper around a Tensorflow back end running the Adam optimizer (Kingma and Ba 2014). All prediction models were trained using an NVIDIA GeForce GTX 1080 Ti GPU. We emphasize the importance of using graphical processing units (GPUs) for both the training of our models and their application within our heuristics. For example, we found that using a GPU resulted in a \(\sim \)25 times speedup during training for the Test Settings from Sect. 4. In addition, in the search stage of both the PS and PSRH heuristics, we are able to create large batches of networks that could be evaluated on a GPU, which greatly increased performance especially for the larger networks in the radiation therapy case study. For training, we used a root mean square error loss function.
Appendix B: PS and PSRH parameter analysis
The following figure illustrates the performance of the PS and PSRH heuristics with varying batch and search size parameters. Each panel is a different test setting and search size. The lines in each plot correspond to a different heuristic and batch size. While there may be performance improvements in going to larger batch and search sizes, generally we see that batch and search sizes of 2 perform quite well. Plus, having smaller batch and search sizes is less computationally expensive. Thus, we use batch and search sizes of 2 in our numerical experiments (Fig. 15).
Appendix C: Treatment network designs
Treatment network designs discussed in Sect. 5 are included here (Figs. 16, 17, 18, 19, 20, 21, 22).
Rights and permissions
About this article
Cite this article
Chan, T.C.Y., Letourneau, D. & Potter, B.G. Sparse flexible design: a machine learning approach. Flex Serv Manuf J 34, 1066–1116 (2022). https://doi.org/10.1007/s10696-021-09439-2
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10696-021-09439-2