Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD11

2 - A2BCD an Asynchronous Accelerated Block Coordinate Descent Algorithm with Optimal Complexity Robert Hannah, UCLA, Los Angeles, CA, 90025, United States, Wotao Yin In this talk we present work on A2BCD, an asynchronous accelerated randomized coordinate descent algorithm. For asynchronous delays that are not too large, we prove that A2BCD has the same complexity as NU_ACDM, the current state-of- the-art coordinate descent algorithm. We then show that this complexity is essentially optimal: That is, it cannot be improved except by a small constant factor without very different algorithmic assumptions. To motivate and clarify our proof techniques, we also present an ODE which is the continuous-time limit of our algorithm and converges to a solution at the same rate. 3 - Conditional Gradient Methods for Stochastic Submodular Maximization Hamed Hassani, University of Pennsylvania, 3330 Walnut Street, Philadelphia, PA, 19104, United States In this talk, we study the problem of constrained and stochastic submodular maximization (both discrete and continuous) and develop gradient methods that achieve a tight approximation guarantee. More precisely, for a monotone and continuous DR-submodular function and subject to a general convex body constraint, we provide stochastic and conditional gradient methods that achieve a (1-1/e)-OPT guarantee (in expectation). By using stochastic continuous optimization as an interface, we also provide the first (1-1/e)-OPT tight approximation guarantee for maximizing a monotone but stochastic submodular set function subject to a general matroid constraint. 4 - Escaping Saddle Points in Constrained Optimization Aryan Mokhtari, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Room 32-D608, Cambridge, MA, 02139, United States, Asuman Ozdaglar, Ali Jadbabaie In this talk, we focus on escaping from saddle points in smooth nonconvex optimization problems subject to a convex set. We propose a generic framework that yields convergence to a second-order stationary point of the problem if the convex set is simple for a quadratic objective function. To be more precise, our results hold if one can find a ?-approximate solution of a quadratic program subject to the convex constraint in polynomial time, where ?<1 is a positive constant that depends on the structure of the convex set. Joint Session MSOM/RMP/Practice Curated: Online Retailing and Marketplace Analytics Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Yun Fong Lim, Singapore Management University, Singapore, 178899, Singapore 1 - The Impact of Repeat Business and Word of Mouth for an E-Retailer Opher Baron, University of Toronto, 105 St George Street, Toronto, ON, M5S 1L7, Canada, Simai He, Hongsong Yuan We use data from the M&SOM data competition to investigate repeat purchase behavior and the impact of word of mouth (WOM) on reference prices at the giant Chinese e-retailer, Taobao. We develop algorithms to study both questions and demonstrate their usage. Our main findings are that improving repeat business is extremely important to Taobao and that it is important to consider WOM affect in the presence of reference price effects. 2 - Urban Consolidation Center or Peer-to-peer Platform? The Solution to Urban Last-mile Delivery Qiyuan Deng, Singapore Management University, Singapore, Xin Fang, Yun Fong Lim The growth of last-mile delivery to urban areas creates negative impacts on the environmental, social, and economic well-being of cities. As a potential solution to address this challenge, an urban consolidation center (UCC) bundles shipments from multiple carriers before delivering them to a city center. Despite the potential benefits, the success rate of UCC projects in practice is low. More recently, a notable number of peer-to-peer platforms have been established to share the carriers’ truck capacity. These capacity sharing platforms can potentially be more economically sustainable, while making urban last-mile delivery more operationally efficient. n TD10 North Bldg 125A

3 - Distributionally Robust Inventory Management with Network Flows - An Application to Ecommerce and Omnichannel Retailing Aravind Govindarajan, Ross School of Business, University of Michigan, Ann Arbor, MI, 48104, United States, Amitabh Sinha, Joline Uichanco A fundamental assumption in network inventory planning is that the underlying joint distribution of demands across locations is known. However, this may not be the case for ecommerce demand due to high volatility in online customer behavior. We propose propose a distributionally robust model for network inventory optimization with reactive recourse, where the worst-case expected cost is minimized over the set of demand distributions satisfying known mean and covariance information. We derive closed-form optimal inventory levels and worst-case distribution for two locations and develop tractable upper bounds and heuristic solutions for the multi-location problem. 4 - Matching Supply with Demand for Online Retailing Yun Fong Lim, Singapore Management University, Lee Kong Chian School of Business, Singapore, 178899, Singapore, Song Jiu, Marcus Teck Meng Ang We consider a joint replenishment, allocation, and fulfillment (JRAF) problem over multiple periods for an online retailer. In each period, the retailer determines the replenishment quantity for each product from each supplier and then allocates the inventory to the FCs. After the demand is realized, the retailer chooses the FCs to satisfy it. The retailer’s objective is to minimize the expected total cost. We have developed a two-stage approach based on robust optimization to solve the JRAF problem. A case study with a major apparel online retailer in Asia suggests that the two-stage approach can reduce the retailer’s current cost by 36.73%, demonstrating a significant value of joint optimization. n TD11 North Bldg 125B Stochastic Models for Biomanufacturing Supply Chains Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Ananth Krishnamurthy, University of Wisconsin-Madison, Madison, WI, 53706, United States Co-Chair: Tugce Martagan, Eindhoven University of Technology, Eindhoven, 5611AZ, Netherlands 1 - Contract Structures for Biomanufacturing Projects with Failure Risks Yasemin Limon, University of Wisconsin-Madison, Madison, WI, 53705, United States, Tugce Martagan, Ananth Krishnamurthy Bio-pharmaceutical companies often subcontract projects from their drug development pipeline to smaller biomanufacturers to reduce costs of failure. These projects require multiple sequential steps with significant uncertainty, leading to risks of not meeting client requirements. Currently, contract terms for these projects are negotiated with limited quantification of the risks of failure. We analyze the performance of two contract structures, Fee For Service (FFS) and Price Per Mass (PPM) and assess how these contacts distribute profits and costs of failure. We then propose a new Scout Before Commit (SBC) contract, and study when the SBC contract is likely to yield better performance. 2 - Optimal Harvesting and Replenishment Policies to Reduce Changeovers in Fermentation Processes Yesim Koca, Eindhoven University of Technology, Eindhoven, Netherlands, Tugce Martagan, Lisa M. Maillart, Ivo Adan We develop a time-based and a condition-based stochastic model to reduce the number of changeovers in the fermentation processes. We determine the optimal harvesting and replenishment policies that maximize the expected total profit obtained from a batch. We analyze the structural characteristics of optimal policies, and illustrate the industry use of the models through a case study. 3 - Biomanufacturing Companies Accelerate Growth Using Operations Research Ananth Krishnamurthy, University of Wisconsin-Madison, 1513 University Avenue, Madison, WI, 53706, United States, Tugce Martagan, Yasemin Limon In the biomanufacturing industry, production and planning decisions are challenging due to batch-to-batch variability and uncertainty in the production yield, quality, cost, and lead times. A multidisciplinary team of researchers developed decision support tools that use a data-driven, operations research- based approach to improve biomanufacturing efficiency. Optimization tools for fermentation and purification operations provide a decision support mechanism that links the underlying biological and chemical processes with business risks and financial trade-offs. Interactive scheduling and capacity planning tools enable efficient use of the expensive and limited resources.

349