INFORMS 2021 Program Book

INFORMS Anaheim 2021

SB30

an assortment family that maximizes the expected revenue of the matching. We show several provable guarantees for the general model, which in particular, significantly improve the approximation factors previously obtained. 3 - Adaptive Bin Packing with Overflow Sebastian Perez Salazar, Georgia Institute of Technology, Atlanta, GA, 30309-4245, United States Driven by the allocation of VMs into servers, we consider the online problem of packing items with random sizes into unit-capacity bins. Upon an item’s arrival, it’s actual size is unknown and only its probabilistic information is available to us. We must irrevocably pack the item into an available bin or pack it in a new bin. After this, we observe the item’s size, and a bin overflow can occur. An overflow incurs a penalty cost and renders the corresponding bin unusable. The goal is to minimize the expected cost given by the sum of the number of opened bins and the overflow penalty. We give an algorithm with expected cost at most a constant factor times the cost incurred by the optimal packing policy when item sizes comes from an i.i.d. sequence. SB30 CC Room 207D In Person: Deep Learning for Quality Assurance in Manufacturing Systems General Session Chair: Xiaowei Yue, Virginia Tech, Blacksburg, VA, 24061, United States Chair: Abdallah A Chehade, University of Michigan-Dearborn, Dearborn, MI, 48128-2406, United States 1 - Deviation-aware Segmentation and Active Landmarks Selection for 3d Printing Weizhi Lin, University of Southern California, Los Angeles, CA, United States, Qiang Huang In 3D printing, geometric quality control demands a reliable deviation representation. The characterization of shape deviation requires the non-rigid registration between the designed and printed products. Manual landmark detection is usually the first step to find this registration, especially for complicated shapes like teeth. To increase the efficiency, we present an automatic landmark selection method in this work. By integrating the geometric properties, a 3D shape will be first segmented via a novel density-based geodesic clustering method. Segment-wise landmarks selection is achieved through an active Gaussian process to ensure most of the landmarks’ geometric information. 2 - Generalized Additive Models For Prediction and Compensation of Shape Deviation of Large-scale Additive Manufactured Products Cesar Ruiz, University of Southern California-Los Angeles, CA, 90007, United States, Davoud Jafari, Tom Vaneker, Qiang Huang Wire and arc additive manufacturing (WAAM) has become an increasingly economically viable way to manufacture components made of high-value materials. WAAM provides an effective method for applications involving large near-net parts with short lead times and millimeter resolution such as the aerospace and oil industries. Due to high residual stresses, current WAAM technologies have poor shape accuracy and high surface roughness, which limit the wide use of this technology in the industry. We propose a tensor product bases expansion to model both the low-order shape distortion and high-order roughness of the manufactured parts. The proposed model enables the optimal compensation of the design of the part to minimize shape distortion. Experimental validation on hollow cylinders shows the effectiveness of the proposed framework. 3 - A Collaborative Gaussian Process Regression Model for Transfer Learning of Capacity Trends Between LI-ION Battery Cells Abdallah A. Chehade, University of Michigan-Dearborn, Hpec,Dearborn, Dearborn, MI, 48128-2406, United States, Ala Hussein A transfer learning method is proposed for forecasting the capacity of lithium-ion battery cells. The proposed approach uses the multi-output Gaussian process regression framework to collaboratively model multiple battery cells. Besides the high prediction accuracy and robustness of the proposed method, it provides uncertainty information, and it has the capability to cross-correlate capacity trends between different battery cells. These two merits make the proposed method a very reliable and practical solution for applications that use battery cell packs with a large number of interconnected battery cells. The proposed method is derived, verified, and compared to benchmark methods on three experimental lithium-ion battery cell datasets. The results show the effectiveness of the proposed method.

SB28 CC Room 207B In Person: Exploiting Structure in Zeroth-order Optimization General Session Chair: Daniel Mckenzie, University of California, Los Angeles, United States Chair: HanQin Cai, University of California, Los Angeles, United States 1 - Curvature-Aware Derivative Free Optimization Bumsu Kim, University of California-Los Angeles, Los Angeles, CA, United States In this work, we present new algorithms for derivative-free optimization which exploit approximate curvature information. The first algorithm, coined Curvature-Aware Random Search (CARS), uses an estimate of the curvature along a search direction to approximate the optimal step size in the given direction. Furthermore, we propose a novel stochastic estimator of the Hessian inverse to construct a new search direction whose expectation is parallel to the Newton vector. This estimator is employed by our second algorithm, coined Stochastic Hessian Inverse Projected Search (SHIPS), to yield a derivative-free approximation to Newton’s method. We benchmark CARS and SHIPS on the MuJoCo control problems and the adversarial attack. The numerical results compare favorably with the other state-of-the-art methods. 2 - Zeroth-order Regularized Optimization (ZORO): Approximately Sparse Gradients and Adaptive Sampling HanQin Cai, University of California-Los Angeles, Los Angeles, CA, United States We consider the problem of minimizing a high-dimensional objective function, which may include a regularization term, using only (possibly noisy) evaluations of the function. Such optimization is also called derivative-free, zeroth-order, or black-box optimization. We propose a new Zeroth- Order Regularized Optimization method, dubbed ZORO. When the underlying gradient is approximately sparse at an iterate, ZORO needs very few objective function evaluations to obtain a new iterate that decreases the objective function. Under a novel approximately sparse gradient assumption and various different convex settings, we show the convergence rate of ZORO is only logarithmically dependent on the problem dimension. Numerical experiments show that ZORO outperforms existing methods with similar assumptions, on both synthetic and real datasets. SB29 CC Room 207C In Person: Advances in Resource Allocation under Uncertainty General Session Chair: Sebastian Perez-Salazar, Georgia Institute of Technology, GA, United States Chair: Alfredo Torrico, Polytechnique Montreal, Montreal, QC, H2V 4G9, Canada 1 - Order Fulfillment under Pick Failure in Omnichannel Ship-from-store Programs Sagnik Das, Carnegie Mellon University, Pittsburgh, PA, United States, R. Ravi, Srinath Sridhar We consider the order fulfillment problem in omnichannel retailing, where in- store and online demand channels cause inventory inaccuracy leading to pick failure at stores. We propose order fulfillment models for every sparse/dense combination of online and in-store demands to optimize labor, shipping, cancellation, and lost-sales costs while accounting for pick failure at stores. We establish structural results for our models and exploit them to optimize over fulfillment policies efficiently. We demonstrate the value of modeling pick failure on data from our collaborating solutions provider to top North American omnichannel retailers. 2 - Multiagent Assortment Optimization in Sequential Matching Markets Alfredo Torrico, Polytechnique Montreal, Montreal, QC, H2V 4G9, Canada, Margarida Carvalho, Andrea Lodi We study a general version of the two-sided sequential matching model. The setting is the following: we (the platform) offer a menu of suppliers to each consumer. Then, every consumer selects, simultaneously and independently, to match with a supplier or to remain unmatched. Suppliers observe the subset of consumers that selected them,and choose either to match a consumer or leave the system. Finally, a match takes place if both the consumer and the supplier sequentially select each other. Each agent’s behavior is probabilistic and determined by a regular discrete choice model. Our objective is to choose

9