Latest Seminars

Frozen-State Approximate Dynamic Programming for Fast-Slow MDPs
Dr Daniel Jiang, University of Pittsburgh

In this talk, we consider infinite horizon Markov decision processes (MDPs) with "fast-slow" structure, meaning that certain parts of the state space move "fast" (and are more influential) while other parts of the state space transition more slowly (and are less influential). Examples of this type of structure arise in a number of practical applications: multi-product inventory control and pricing, machine maintenance, multi-class queueing, and energy demand response. We propose an approximate value iteration algorithm based on the idea of periodically "freezing" the slow states, solving a set of simpler finite-horizon MDPs, and applying value iteration to an auxiliary MDP that transitions on a slower timescale (and smaller discount factor). We present analyses of the regret of policies generated by our approach, along with empirical results that demonstrate its computational benefits.

Date 13.05.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

Emission Reduction through Regulating Indirect Sources (joint work Luyi Gui and Sai Zhao)
Dr Shiliang (John) Cui, Georgetown University

Emission from diesel trucks such as Nitrogen Oxides causes severe air pollution. However, direct regulation on trucking companies for their use of diesel trucks typically falls out of the jurisdiction of local governments. A legislative alternative is to regulate other sectors that prompt diesel truck usage in the local region, called indirect emission sources. The first of such regulations is Southern California's Rule 2305, the Warehouse Indirect Source Rule (ISR). Passed in May 2021, the ISR holds local warehouses responsible for the diesel truck trips to their facilities through a mitigation fee. The goal of the ISR is to incentivize warehouses to hire electric semi-trucks to improve air quality and thus public health. Motivated by this new policy, we explore the environmental impact of the ISR and the industry burden that it introduces, compared to a hypothetical direct source rule (DSR) that regulates trucking companies. We find that ISR can indeed lead to higher adoption of electric semi-trucks than DSR, especially when the mitigation fees for diesel truck trips are small. However, using the mitigation fee collected from warehouses to subsidize trucking companies’ electric semi-truck investments, a current practice of the ISR, can backfire and reduce industry adoption of electric semi-trucks. Interestingly, depending on the distribution of truck trips' distances, a higher mitigation fee for using diesel trucks can also lead to lower adoption of electric semi-trucks. Finally, we explore the practical implications of the ISR using real warehouse data from Southern California.  

Date 06.05.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

Intertemporal Price Discrimination via Randomized Promotions
Dr Jiahua Wu, Imperial College Business School

The undesirable but inevitable consequence of running promotions is that consumers can be trained to time their purchases strategically. In this paper, we study randomized promotions, where the firm randomly offers discounts over time, as an alternative strategy of intertemporal price discrimination. Specifically, we consider a base model where a monopolist sells a single product to a market with a constant stream of two market segments. The segments are heterogeneous in both their product valuations and patience levels. The firm pre-commits to a price distribution, and in each period, a price is randomly drawn from the chosen distribution. We characterize the optimal price distribution as a randomized promotion policy and show that it serves as an intertemporal price discrimination mechanism such that high-valuation customers would purchase immediately at a regular price upon arrival and low-valuation customers would wait for a random promotion. Compared against the optimal cyclic pricing policy, which is optimal within the strategy space of all deterministic pricing policies, the optimal randomized pricing policy beats the optimal cyclic pricing policy if low-valuation customers are sufficiently patient and the absolute discrepancy between high and low customer valuations is large enough. We extend the model in two directions. We first consider Markovian pricing policies where prices are allowed to be intertemporally correlated in a Markovian fashion. This additional maneuver allows the firm to reap an even higher profit when low-valuation customers are sufficiently patient, by avoiding consecutive promotions but on average running the promotion more frequently with a smaller discount size. We then consider a model with multiple customer segments, and show that a two-point price distribution remains optimal and our conclusion from the two-segment base model still holds under certain conditions that are adopted in the literature. Our results imply that the firm may want to deliberately randomize promotions in the presence of forward-looking customers.

Date 29.04.2022
Time 4:00 - 5:15 PM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

Joint Statistics Seminar - Community Detection on Mixture Multi-layer Networks via Regularized Tensor Decomposition
Prof. Dong XIA, Department of Mathematics, HKUST

We study the problem of community detection in multi-layer networks, where pairs of nodes can be related in multiple modalities.  We introduce a general framework, i.e., mixture multi-layer stochastic block model (MMSBM), which includes many earlier models as special cases.  We propose a tensor-based algorithm (TWIST) to reveal both global/local memberships of nodes, and memberships of layers.  We show that the TWIST procedure can accurately detect the communities with small misclassification error as the number of nodes and/or the number of layers increases. Numerical studies confirm our theoretical findings.  To our best knowledge, this is the first systematic study on the mixture multi-layer networks using tensor decomposition.  The method is applied to two real datasets: worldwide trading networks and malaria parasite genes networks, yielding new and interesting findings. 

Based on joint work with Bing-Yi Jing, Ting Li and Zhongyuan Ly.

Date 22.04.2022
Time 11:00 am - 12:00 noon
Venue Zoom ID 920 0082 3966 (Passcode: STAT)

Data-pooling Reinforcement Learning for Personalized Healthcare Intervention
Prof Pengyi Shi, Purdue University

Personalized intervention management in healthcare has received a rapidly growing interest in the big-data era yet still is a burgeoning field. A key challenge for personalization in healthcare is data scarcity. This small sample issue makes standard learning methods hard to learn the right policy and/or suffer from large variances. In this research, we extend the data-pooling technique from the bandit setting to the reinforcement learning (RL) context. RL models explicitly account for future cost/reward and are more suitable for healthcare management problems. We develop a novel data-pooling estimator in the RL context, and establish theoretical performance guarantee for RL with data-pooling. We demonstrate its empirical success on a real hospital dataset with an application to reduce 30-day hospital readmission rate. This is a joint work with Xinyun Chen and Xiuwen Wang from CUHK Shenzhen.

Date 22.04.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

Randomized FIFO Mechanism
Dr Chiwei Yan, The University of Washington Seattle

We study the matching of jobs to workers in a queue, e.g. a ridesharing platform dispatching drivers to pick up riders at an airport. Under FIFO dispatching, the heterogeneity in trip earnings incentivizes drivers to cherry-pick, increasing riders' waiting time for a match and resulting in a loss of efficiency and reliability. We first present the direct FIFO mechanism, which offers lower-earning trips to drivers further down the queue. The option to skip the rest of the line incentivizes drivers to accept all dispatches, but the mechanism would be considered unfair since drivers closer to the head of the queue may have lower priority for trips to certain destinations. To avoid the use of unfair dispatch rules, we introduce a family of randomized FIFO mechanisms, which send declined trips gradually down the queue in a randomized manner. We prove that a randomized FIFO mechanism achieves the first best throughput and the second best revenue in equilibrium. Extensive counterfactual simulations using data from the City of Chicago demonstrate substantial improvements of revenue and throughput, highlighting the effectiveness of using waiting times to align incentives and reduce the variability in driver earnings.

 

Joint work with Francisco Castro, Hongyao Ma and Hamid Nazerzadeh

Date 08.04.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

An Asymptotically Tight Learning Algorithm for Mobile-Promotion Platforms
Prof Anyan Qi, The University of Texas at Dallas

Operating under both supply-side and demand-side uncertainties, a mobile-promotion platform conducts advertising campaigns for individual advertisers. Campaigns arrive dynamically over time, which is divided into seasons; each campaign requires the platform to deliver a target number of mobile impressions from a desired set of locations over a desired time interval. The platform fulfills these campaigns by procuring impressions from publishers, who supply advertising space on apps, via real-time bidding on ad exchanges. Each location is characterized by its win curve, i.e., the relationship between the bid price and the probability of winning an impression at that bid. The win curves at the various locations of interest are initially unknown to the platform, and it learns them on the fly based on the bids it places to win impressions and the realized outcomes. Each acquired impression is allocated to one of the ongoing campaigns.  The platform's objective is to minimize its total cost (the amount spent in procuring impressions and the penalty incurred due to unmet targets of the campaigns) over the time horizon of interest.

Our main result is a bidding and allocation policy for this problem. We show that our policy is the best possible (asymptotically tight) for the problem using the notion of regret under a policy, namely the difference between the expected total cost under that policy and the optimal cost for the clairvoyant problem (i.e., one in which the platform has full information about the win curves at all the locations in advance): The regret under any policy is , where  is the number of seasons, and that under our policy is . We demonstrate the performance of our policy through numerical experiments on a test bed of instances whose input parameters are based on our observations at a real-world mobile-promotion platform.

Date 18.03.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 978 2332 0242 (passcode 370129)

Corporate Social Responsibility in Supply Chain: Green or Greenwashing?
Prof Jing Wu, The Chinese University of Hong Kong (CUHK) Business School

Perception regarding a focal firm's corporate social responsibility (CSR) depends not only on itself but also on its known suppliers. This paper provides the first empirical evidence linking CSR and supply chain information disclosure together. We uncover robust evidence that listed firms voluntarily disclose environmentally responsible suppliers while selectively not disclosing" bad" ones, effectively greenwashing their supply chain image. This selective disclosure of green suppliers is prevalent among listed firms across the world. Such corporate behavior is increasing in public awareness of climate change, decreasing in regulations on CSR information transparency. It is more salient for firms who face higher competition or care more about their brand awareness, and for firms that are more profit-driven or held more by institutional investors. Firms that greenwash supply chains observe an increase in sales and valuation, suggesting that consumers and investors do not fully take greenwashing of listed firms into account.

Date 11.03.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 931 9977 8233 (passcode 792828)

Facility Location with Competition or Decision-dependent Uncertainty: Models, Algorithms and Extensions
Prof Siqian Shen, University of Michigan at Ann Arbor

Facility location models are ubiquitously involved in modern transportation and logistics problems. We present recent results of two types of facility-location models that involve (i) competition and probabilistic customer choice or (ii) location-dependent uncertain demand with ambiguously known distribution. For (i), we study a Stackelberg game that admits a bilevel mixed-integer nonlinear program (MINLP) formulation, and derive an equivalent, single-level MINLP reformulation and exploit the problem structures to derive valid inequalities, based on submodularity and concave overestimation, respectively. We also study various model extensions by considering general facility setup costs, multiple competitors, as well as other types of decisions for planning facilities. We conduct numerical studies to demonstrate that the exact algorithm significantly accelerates the computation of CFLP on large-sized instances that have not been solved optimally or even heuristically by existing methods. For (ii), we represent moment information of stochastic demand as piecewise linear functions of location decisions, and then develop an exact mixed-integer linear programming reformulation of a decision-dependent distributionally robust optimization model. Our results draw attention to the need of considering various impacts of competition and location choices on customer demand during strategic facility planning. (Papers related to the talk: https://arxiv.org/abs/2103.04259 ; https://www.sciencedirect.com/science/article/abs/pii/S0377221720309449)

Date 04.03.2022
Time 10:30 - 11:45 AM
Venue Zoom ID: 976 9630 8456 (passcode 108728)

Joint Statistics Seminar - Exact Simulation of Generalized Gamma Process and Its Application in Caron-Fox Random Graph
Dr. Junyi ZHANG, The Hong Kong Polytechnic University

Generalized Gamma process is a pure-jump subordinator with infinite activity, it can be used to construct a flexible two-parameter complete random measure, whose application has appeared in various areas. In this talk, we present an exact simulation algorithm to sample from the largest n jumps of a generalized Gamma process. The algorithm immediately implies a method to sample from the celebrated Poisson-Dirichlet distribution, we will illustrate this method with numerical examples. As an application of our algorithm, we review the construction of the Caron-Fox random graph and discuss a potential modification to its simulation algorithm.

Date 25.02.2022
Time 11:00 am - 12:00 noon
Venue Zoom ID 920 0082 3966 (Passcode: STAT)

Nudging Patient Choice by Messaging
Miss Jiayi Liu, Emory University Goizueta Business School

Patient no-shows for scheduled medical appointments are of great concern for many health care providers. In this paper, we tackle the no-show problem by applying insights from behavioral science. Specifically, we ''nudge'' patients into arriving for their scheduled appointment using text reminders of their upcoming visit. We conduct a field experiment at an outpatient specialty clinic where we add to the standard message, an additional line of text that indicates a potentially long wait for the next available appointment (we call this intervention ''waits framing''). Based on a difference-in-differences estimation strategy, we find that waits framing messaging significantly reduced no-shows by a factor of 28.6%. In addition, we find that patients with greater sensitivity to wait, such as those with urgent conditions and those willing to select unpopular slots, are more responsive to the nudge. Through a laboratory experiment, we uncover the mechanism that underlies the nudge---waits framing serves to trigger loss- aversion pertaining to the individual's position in the queue, thereby increasing the perceived cost of missing an appointment. Through the combination of field and designed lab studies, we provide both external and internal validity to the effects of waits framing, and identify the underlying mechanism and heterogeneity in response. Our results have significant implications for clinical operations. At the study site, the resulting improvement in capacity utilization and patient throughput led to a 5.2% increase in clinic revenue. Our findings contribute to the literature on behavioral queuing by showing that through appropriately framed messages, queue operators can tap into the behavioral biases of individuals in order to engender a desired queuing response such as a reduction in queue abandonment.

Date 16.12.2021
Time 10:00 - 11:15 AM
Venue Zoom ID: 940 1790 0792 (passcode 350615)

Avoiding Fields on Fire: Information Dissemination Policies for Environmentally Safe Crop-Residue Management
Dr Mehdi Farahani, Massachusetts Institute of Technology

Agricultural open burning, i.e., the practice of burning crop residue in harvested fields to prepare land for sowing a new crop, is well-recognized as a significant contributor to CO2 and black-carbon emissions, and long-term climate change. Low-soil-tillage practices using a specific agricultural machine called Happy Seeder, which can sow the new seed without removing the previous crop residue, have emerged as the most effective and profitable alternative to open burning. However, given the limited number of Happy Seeders that the government can supply, and the fact that farmers incur a significant yield loss if they delay sowing the new crop, farmers are often unwilling to wait to be processed by the Happy Seeder and, instead, decide to burn their crop residue. We study how the government can use effective information-disclosure policies in the operation of Happy Seeders to minimize agricultural open burning. A Happy Seeder is assigned to process a group of farms in an arbitrary order. The government knows, but does not necessarily disclose, the Happy Seeder’s schedule at the start of the sowing season. Farmers incur a disutility per unit of time while waiting for the Happy Seeder due to the yield loss as a result of late sowing of the new crop. If the Happy Seeder processes a farm, then the farmer gains a positive utility. At the beginning of each period, each farmer decides whether to burn her crop residue or to wait, given the information provided by the government about the Happy Seeder’s schedule. We propose a class of information-disclosure policies, which we refer to as dilatory policies that provide no information to the farmers about the schedule until a pre-specified period and then reveal the entire schedule. By obtaining the unique symmetric Markov perfect equilibrium under any dilatory policy, we show that the use of an optimal dilatory policy can significantly lower the number of farms burnt compared to that under the full- disclosure and the no-disclosure policies. Using data from the rice-wheat crop system in northwestern India – an area of the world with the highest prevalence of open burning – we conduct a comprehensive case study and demonstrate that the optimal dilatory policy can reduce CO2 and black-carbon emissions by at least 14%.

Date 13.12.2021
Time 10:00 - 11:15 AM
Venue Zoom ID: 915 8229 8109 (passcode 162225)

Deep Reinforcement Learning for Sequential Targeting
Ms. Wen Wang, Carnegie Mellon University

Deep reinforcement learning (DRL) has opened up many unprecedented opportunities in revolutionizing the digital marketing field. In this study, we designed a DRL-based personalized targeting strategy. We show that the strategy is able to address four important challenges in this area. 1) Sequential-decisions: accounting for the dynamic sequential behavior of consumers; 2) Forward-looking: balancing between a firm’s current revenue and future revenues; 3) Earningwhile-learning: maximizing profits while continuously learning through exploration-exploitation; 4) Scalability: coping with a high-dimensional state and policy space. We illustrate the above through a novel design of a DRL-based artificial intelligence (AI) agent. Further, in order to better understand the potential underlying mechanisms, we conducted multiple interpretability analyses to explain the patterns of learned optimal policy at both the individual and population levels. Our findings provide important managerial-relevant and theory-consistent insights. For instance, consecutive price promotions at the beginning can capture price-sensitive consumers’ immediate attention, while carefully spaced non-promotional “cool-down” periods between price promotions can allow consumers to adjust their reference points. Besides, consideration of future revenues is necessary from a long-term horizon, but weighing the future too much can also dampen revenues. In addition, analyses of heterogeneous treatment effects suggest that the optimal promotion sequence pattern highly varies across the consumer engagement stages. Overall, our study results demonstrate DRL’s potential to optimize these strategies’ combination to maximize long-term revenues.

Date 13.12.2021
Time 9:30am - 11:00am (Hong Kong Time)
Venue Zoom ID: 954 7599 5651 (Passcode:246467)

Supply Chain Visibility: Impact and Value of Real-time Resource Allocation
Dr Guodong Lyu, National University of Singapore

In recent years, we have seen a surge of interest in supply chain visibility. Under this paradigm, decision- makers are able to trace the real-time data (e.g., stock level, resource allocation flow) along the entire supply chain so that they can identify the decision-making bottlenecks and take actions more efficiently. Motivated by the Gaze Heuristic, we propose a target-based online planning framework to deal with real-time resource allocation problems in both stationary and nonstationary environments. Leveraging on the Blackwell's Approachability Theorem and Online Convex Optimization tools, we characterize the near-optimal performance guarantee of our online solution in comparison with the offline optimal solution, and explore the properties of different allocation policies.

We use synthetic and real data from various industries, from supply chain planning in manufacturing, to resource deployment in ride-sharing markets, to examine the impact and value of these real-time solutions in practice: (1) we present a new insight into the impact of supply chain visibility on the capacity configuration in the capacity pooling system. Our results show that the pooling system does not need to hold any safety stock to deliver the required demand fulfillment service if real-time allocation with full visibility is utilized, when the number of customers is sufficiently large in the system; (2) we study a real-time ride-matching problem in the ride-sourcing context, with multi-objectives (e.g., service quality, revenue) to be considered. We develop a new technique that can be used to choose the weight adaptively over time, based on real-time tracking of the gaps in attained performance and a set of performance targets. Our results show that the real-time matching policy could potentially contribute to the long-term sustainability and reputation of the ride-sourcing platform by dispatching more orders to drivers with higher service quality, without sacrificing the short-term platform revenue.

Date 10.12.2021
Time 10:00 - 11:15 AM
Venue Zoom ID: 948 3060 9615 (passcode 646201)

Video Game Analytics
Mr Xiao Lei, Columbia University

Video games represent the largest and fastest-growing segment of the entertainment industry, which involves 3 billion gamers and garners $180 billion annually.  Despite its popularity in practice, it has received limited attention from the operations community. Managing product monetization and engagement presents unique challenges due to the characteristics of gaming platforms, where players and the gaming platform have repeated (and endogenously controlled) interactions. In this talk, we describe a body of work that provides the first analytical results for this emerging market. In the first part, we discuss a prevailing selling mechanism in online gaming known as a loot box. A loot box can be viewed as a random bundle of virtual items, whose contents are not revealed until after purchase. We consider how to optimally price and design loot boxes from the perspective of a revenue-maximizing video game company, and provide insights on customer surplus and protection under such selling strategies. In the second part, we consider how to manage player engagement in a game where players are repeatedly matched to compete against one another. Players have different skill levels which affect the outcomes of matches, and the win-loss record influences their willingness to remain engaged. Leveraging optimization and real data, we provide insights on how engagement may increase with optimal matching policies, adding AI bots, and providing a pay-to-win feature.

Date 09.12.2021
Time 10:00 - 11:15 AM
Venue Zoom ID: 937 4443 9854 (passcode 923966)