Informs Annual Meeting 2017

TA18

INFORMS Houston – 2017

3 - Difficulties of Nascent Entrepreneurs: What Prevents Individuals from Starting a New Venture? Jessica Moser, Assistant Lecturer and PhD Candidate, Technical University of Dortmund, Vogelpothsweg 87, Dortmund, 44227, Germany, Jessica.Moser@udo.edu This research analyzes the influences of multiple barriers nascent entrepreneurs (NEs) encounter on their outcome status - still trying, disengaged or started - and on their start-up intention after disengaging. Drawing upon a novel and recent sample of 872 NEs, this paper finds that recently disengaged individuals have faced significantly more barriers than other NEs. A deeper analysis displays that financing difficulties and especially personal barriers hinder a venture creation. In addition, barriers beyond a NE’s own control are identified to be lethal for the continuance of the intention to create a new business. Major implications for NEs and policymakers regarding barriers are derived. 4 - Dimension of VC-E Trust Based on the “Guanxi” Culture in China li haiyan, Harbin Engineering University, Harbin, 000451, China, 643076746@qq.com, Yang hongtao The dimension of VC - E trust is divided into three aspects: computational trust, knowledgeable trust and relational trust. The results show that relationship trust is particularly important to the performance of start-ups in China. VC and E can share resources in the phase of relational trust.Acknowledgement: This paper is funded by the International Exchange Program of Harbin Engineering University for Innovation-oriented Talents Cultivation. 340A Random Graphs and Learning in Applied Probability Sponsored: Applied Probability Sponsored Session Chair: Jiaming Xu, Purdue University, West Lafayette, IN, 47907, United States, xu972@purdue.edu 1 - The Lovasz Theta Function of Random Graphs and Community Detection in the Hard Regime Cris Moore, Santa Fe Institute, Santa Fe, NM, United States, moore@santafe.edu Community detection and graph coloring is believed to have a hard regime, in between an informational threshold and a computational one, where inference is possible but exponentially hard. In particular, in this range it should be hard to refute the existence of a community structure in a null random graph model. We confirm this for refutations that use the Lovasz theta function, or equivalently degree-two sum-of-squares proofs. This is joint work with Jess Banks and Robert Kleinberg. 2 - Estimating the Number of Connected Components of Large Graphs Based on Subgraph Sampling Yihong Wu, Yale University, yihong.wu@yale.edu, Jason Klusowski Learning properties of large graphs from samples is an important problem in statistical network analysis. We revisit the problem of Frank ‘78 of estimating the numbers of connected components based on subgraph sampling, where a subgraph induced by vertices drawn uniformly at random is observed. The key question is whether accurate estimation is possible if we only sample a vanishing fraction of the vertices. We show that this is impossible if the graph contains high- degree vertices or long induced cycles; otherwise, it is possible by accessing only sublinear number of vertices. Optimal sample complexity are obtained for several classes of graphs including forests, cliques and chordal graphs. 3 - Optimal Community Recovery on the Weighted Stochastic Block Model Varun Jog, University of Wiscconsin-Madison, vjog@wisc.edu The Stochastic Block Model (SBM), which is one of the most well-studied statistical models for community detection, has an important limitation: it assumes that each network edge is drawn from a Bernoulli distribution. This is rather restrictive, since weighted edges are ubiquitous in scientific applications, and disregarding edge weights results in a loss of information. In this talk, we study a weighted generalization of the SBM. We propose and analyze a novel algorithm for community estimation and show that it is optimal in terms of its rate of convergence. 4 - On the Local Minima of Nonconvex Matrix Completion with Arbitrary Condition Numbers and General Sampling Patterns Xiaodong Li, University of California, Davis, CA, United States, xdgli@ucdavis.edu, Ji Chen In recent years much progress has been made in the theory of nonconvex matrix completion. However, strong conditions are usually required for the condition numbers and sampling patterns. In this work, we study the properties of local minima of nonconvex matrix completion with general sampling patterns and arbitrary condition numbers. We apply this general framework to some specific random graph-based sampling patterns to obtain novel approximate recovery TA17

results for all local minima. As a byproduct, we also improve the state-of-the-art no-spurious-local-minima results in the literature under strong conditions of the condition numbers and sampling patterns.

TA18

340B Joint Session MSOM/APS: Supply Chain and Inventory in Applied Probability II Sponsored: Applied Probability Sponsored Session Chair: Xiting Gong, The Chinese University of Hong Kong, Hong Kong, xtgong@se.cuhk.edu.hk Co-Chair: Xiuli Chao, University of Michigan, Ann Arbor, MI, 48109, United States, xchao@umich.edu 1 - Asymptotic Optimality of Constant Order Policies for Lost Sales Inventory Models with General Supply Functions Jinzhi Bu, The Chinese University of Hong Kong, Hong Kong, njujzhbu@gmail.com, Xiting Gong, Dacheng Yao In this paper, we study the constant-order policies (COP) for the lost-sales inventory system with positive lead times and general supply functions. Besides analyzing the asymptotic properties of the best COP with large lead times and large penalty costs, we construct a simple while asymptotically optimal heuristic COP and carry out extensive numerical studies to investigate the COP’s performances. 2 - Joint Assortment and Pricing Optimization with Demand Learning Sentao Miao, University of Michigan, Ann Arbor, MI, 48104, United States, semiao@umich.edu, Xiuli Chao We study a joint assortment and pricing optimization problem where customers arrive sequentially and make purchasing decisions based on a multi-nomial logit (MNL) model. Without prior information about customers’ demand for products, the retailer selects subset of products for display and set their prices through learning customer behavior to maximize its expected total profit over a finite selling horizon. By a special structure of the MNL model, we design an algorithm that estimates the consumer’s demand and sets the assortment and price for each period. The algorithm balances the trade-o ff between learning the demand and maximizing the revenue, and is shown to be asymptotically optimal. 3 - Optimality of Order-up-to Policies for Serial Inventory Systems with Lost Sales Woonghee Tim Huh, University of British Columbia, Sauder School of Business, Operations and Logistics Division, Vancouver, BC, V6T.1Z2, Canada, tim.huh@sauder.ubc.ca, Marco Bijvank, Ganesh Janakiraman We study how to manage a serial inventory system under periodic review when excess demand at the most downstream stage is lost. First, we show that there is an echelon order-up-to policy that is asymptotically optimal as the lost-sales penalty cost parameter grows large. We also show that this asymptotic optimality result is robust in the sense that there is a large family of order-up-to policies such that it continues to hold. Next, we propose a heuristic rule to set the order-up-to levels based on the algorithm of Clark and Scarf where the resulting policy performs close to the best order-up-to policy even when the penalty cost is not large. Finally, we extend the optimality results to (R,nQ) policies. 4 - Performance Bounds and Asymptotic Optimality of Modified (R,Q) Policies for Stochastic Distribution Inventory Systems Frank Y. Chen, City University of Hong Kong, Dept of Management Sciences, AC1, Tat Chee Avenue, Hong Kong, youhchen@cityu.edu.hk, Han Zhu, Ming Hu, Yi Yang We study a classic one-warehouse multi-retailer distribution system, in which any inventory replenishment at each location incurs a fixed-plus-variable cost and takes a constant lead time. The optimal policy is unknown. We identify an easy- to-compute heuristic policy within the class of modified echelon (r, Q) policies that does not require an integer-ratio property or a synchronized, nested ordering property, yet has easily computable performance bounds. The proposed heuristic policy is asymptotically optimal.

275