Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC09

material handling system. In our study, we simulate different facility layout scenarios to empirically derive a multi-objective function to capture the traffic issue in addition to the distance. Finally, we test the performance of the new objective function by comparing with previous results. 2 - Algorithms for Short-term Demand Supply Matching in Semiconductor Supply Chains Lars Moench, University of Hagen, Enterprise-wide Software Systems, Universitaetsstrasse 1, TGZ, Hagen, 58097, Germany, Raphael Herding, Thomas Ponsignon, Alexander Seitz, Hans Ehm We propose algorithms for short-term demand supply matching in semiconductor supply chains. The algorithms can be used to simultaneously re-promise orders in demand fulfillment processes. The resulting mixed integer programming formulations are solved using time-based decomposition approaches. The algorithms are assessed in a rolling horizon setting including master planning and order promising activities. We compare the obtained results with results for a rule-based order repromising algorithm. The results demonstrate that the novel short-term demand supply matching algorithms outperform the rule-based approaches. 3 - A Review of Product Ramping Models in Semiconductor Manufacturing Atchyuta B. Manda, Doctoral Student, North Carolina State University, Campus Box 7906, Raleigh, NC, 27695-7906, United States, Reha Uzsoy We review the state of the art in modelling the ramping up of new products in semiconductor wafer fabrication facilities. We present a taxonomy of different problem formulations, learning models, capacity models and related yield and process learning, and suggest directions for future research. 4 - Intel Minifab and Observations on Operational Control Leon McGinnis, Georgia Institute of Technology, Atlanta, GA, 30332-0205, United States This famous but aging case study comes to life again to demonstrate some fundamental truths—that our analysis models must clearly distinguish plant and control (which they don’t today), that operational control is implemented through material handling, and that there is no such thing as a “queue” in the factory. If we simulate queues, we are not simulating the factory, and thus cannot evaluate implementable operational control (scheduling) policies. n TC11 North Bldg 125B Stochastic Inventory Control: Asymptotics and Approximations Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Linwei Xin, University of Chicago, Chicago, IL, 60637, United States 1 - Exploiting Random Lead Times in Inventory Systems Alexander Stolyar, University of Illinois at Urbana-Champaign, 1308 W. Main Street, 156CSL, Urbana, IL, 61801, United States, Qiong Wang We study the classical single-item inventory system with random replenishment lead times and order crossovers, and propose a new policy that exploits the lead time randomness. Instead of focusing on the inventory position, our policy uses the net inventory level to set a dynamic target for inventory in-transit, and places orders to follow that target. The policy provides a potentially infinite inventory cost reduction compared with the classical Constant Base Stock (CBS) policy. In the case of exponentially distributed lead times, we prove that, as the demand rate becomes large, the expected (absolute) inventory level under our policy vanishes relatively to that under CBS policy. 2 - LP-based Order-up-To Control for Stochastic Inventory Systems with Sequential Probabilistic Service Level Constraints Linwei Xin, University of Chicago, 5807 S. Woodlawn Avenue, Chicago, IL, 60637, United States, Lai Wei, Stefanus Jasin We consider a stochastic inventory model with non-stationary demands, positive lead time, and sequential probabilistic service level constraints. This is a notoriously difficult problem to solve and, to date, not much progress has been made in understanding the structure of its optimal control, especially for the lost- sales inventory system. In this paper, we propose a simple order-up-to control, whose parameters can be calculated using the optimal solution of a deterministic approximation of the backorder inventory system, and show that it is asymptotically optimal for both the backorder and lost-sales systems in the regime of high service level requirement.

n TC09 North Bldg 124B

Nonlinear and Stochastic Optimization Sponsored: Optimization/Nonlinear Programming Sponsored Session Chair: Raghu Bollapragada, Northwestern University, Evanston, IL, 60201, United States 1 - Randomness and Permutations in Coordinate Descent Mert Gurbuzbalaban, Rutgers, New Brunswick, NJ, United States We consider coordinate descent (CD) methods with exact line search on convex quadratic problems. Our main focus is to study the performance of the CD method that use random permutations in each epoch and compare it to the performance of the CD methods that use deterministic orders and random sampling with replacement. We focus on a class of convex quadratic problems with a diagonally dominant Hessian matrix, for which we show that using random permutations instead of random with-replacement sampling improves the performance of the CD method in the worst-case. This is joint work with Asu Ozdaglar, Nuri Denizcan Vanli and Stephen Wright. 2 - A Progressive Batching L-BFGS Method for Machine Learning Raghu Bollapragada, Northwestern University, 2145 Sheridan We present a new version of the L-BFGS algorithm for machine learning that combines three basic components - progressive batching, a stochastic line search, and stable quasi-Newton updating. The motivation for this approach is to combine the early efficiency and good generalization properties of the SGD method and the fast convergence, stability and parallelization opportunities offered by large batches. We illustrate the performance of the method on logistic regression and deep neural network models. 3 - Inexact Non-convex Newton-type Methods Fred Roosta, University of Queensland, Brisbane, Australia For solving large-scale non-convex problems, we propose inexact variants of trust region and adaptive cubic regularization methods, which incorporate various approximations. In particular, in addition to inexact sub-problem solves, both the Hessian and the gradient are suitably approximated. Using mild conditions on such approximations, we show that our proposed inexact methods achieve similar optimal worst-case iteration complexities as the exact counterparts. Our proposed algorithms do not require knowledge of any unknowable problem-related quantities and are implementable in practice. We also examine the empirical performance of our algorithms on some real datasets. 4 - Random Projections for Faster Non-convex Optimization Mert Pilanci, Stanford University, Stanford, CA, United States Randomized dimension reduction has recently become a powerful tool in machine learning. We consider random projection methods in the context of non- convex optimization problems. First, we introduce a statistical model where the maximum likelihood estimator reduces to fitting a single layer neural network. We show that a second order optimization method with a suitable initialization recovers the global optimum under certain assumptions on the data. We then introduce random projection and sampling strategies that enables faster convergence in terms of clock time. Our results suggest that the proposed method can outperform existing stochastic methods in large scale non-convex optimization. OM in Semiconductor Manufacturing I Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: John W Fowler, Arizona State University, Arizona State University, Tempe, AZ, 85287-4706, United States Chair: Lars Moench, University of Hagen, University of Hagen, Hagen, 58097, Germany 1 - Factory Logistics: Material Handling Performance in a Semiconductor Fab Gwangjae Yu, Arizona State University, AZ, United States, John W. Fowler We study a facility layout problem of a semiconductor manufacturing system (SMS). The objective function in facility layout problems usually is to minimize the material flow distance. However, there is a question if this is still valid for an SMS, which is characterized by re-entrant flow and heavy traffic. As a consequence, an SMS often suffers significant traffic congestion throughout the n TC10 North Bldg 125A Road, Evanston, IL, 60201, United States, Jorge Nocedal, Hao-Jun M. Shi, Dheevatsa Mudigere, Ping T. Peter-Tang

310