Research
Seminars
Filter results | Clear Filters
12:15 PM - 1:30 PM
Lauren Lu
Generative AI in Action: Field Experimental Evidence on Worker Performance in E-Commerce Customer Service Operations
-
Generative AI in Action: Field Experimental Evidence on Worker Performance in E-Commerce Customer Service Operations
Speaker: Lauren Lu
Time: 12:15 PM - 1:30 PM
Location:
12:15 PM - 1:30 PM
Michael Hamilton
Semi-Personalized Pricing: Algorithms and Implications
-
Semi-Personalized Pricing: Algorithms and Implications
Speaker: Michael Hamilton
Time: 12:15 PM - 1:30 PM
Location:
12:00 PM - 1:45 PM
Bijan H. Mazaheri
Synthetic Potential Outcomes and Causal Mixture Identifiability
-
Synthetic Potential Outcomes and Causal Mixture Identifiability
Speaker: Bijan H. Mazaheri
Time: 12:00 PM - 1:45 PM
Location:
12:15 PM - 1:45 PM
Galit Yom-Tov
Operationalizing Emotional Load: The Human Side of Queueing Systems
-
Operationalizing Emotional Load: The Human Side of Queueing Systems
Speaker: Galit Yom-Tov
Time: 12:15 PM - 1:45 PM
Location:
12:15 PM - 1:30 PM
Daniel Guetta
Teaching Business School Students in the 21st Century: data, coding, AI, and more
-
Teaching Business School Students in the 21st Century: data, coding, AI, and more
Speaker: Daniel Guetta
Time: 12:15 PM - 1:30 PM
Location:
The world is changing exponentially faster than it once was, and topics that used to be the preserve of engineers alone are now central to many businesses' core strategy. How should we change the way we teach technical topics in business schools in response? In this talk, I will begin by briefly sharing some lessons I have learned in my seven years at Columbia, during which I have had the privilege of teaching 10 distinct classes, with just under 7,000 students enrolled from Columbia Business School's MBA and EMBA programs, and from our joint programs with the engineering school. The bulk of my talk will then focus on five case studies, time permitting, in which I will discuss the tools and cases I use to teach five specific topics: the Lasso, fundamentals of deep learning, coding in Python, large language model embeddings, and market design. I will end with a discussion of what comes next for us as educators in business schools.
12:15 PM - 1:30 PM
Rad Niazadeh
Dynamic Matching for Refugee Resettlement: A Case Study
-
Dynamic Matching for Refugee Resettlement: A Case Study
Speaker: Rad Niazadeh
Time: 12:15 PM - 1:30 PM
Location:
Refugee resettlement is an international effort that aims to provide a durable solution for the current global refugee crisis. The goal is to help refugee families to find a new home in a host country and eventually find a new job to get “resettled”. In this seminar, I will talk about a recent paper in partnership with a major national agency working on refugee resettlement in the United States. In this work, we re-design the core dynamic matching algorithm used by our partner, for sequential yearly assignment of refugee cases to our partner's affiliate locations. These localities should be thought of as service centers providing vocational services or assistance with job search, and many times are short in staff. I discuss various operational intricacies in this dynamic matching problem, such as lack of reliable arrival prior data, predicting employment outcomes of each match, and controlling backlogs in those service centers. I also discuss regulatory constraints imposed on the problem, such as family re-unification ties for refugees and their implications on our algorithm. Then I will introduce a new algorithmic framework to study this problem, through which I show how to design and analyze near-optimal learning-based primal-dual algorithms that aim to maximize employment outcomes while respecting operational and regulatory constraints in this problem. Time permits, I'll discuss a case study for evaluating the empirical performance of our algorithms using our partner's data and discuss some details of our collaboration.
12:15 PM - 1:30 PM
Canan Gunes Corlu
Uncertainty Quantification in Digital Twin Simulations
-
Uncertainty Quantification in Digital Twin Simulations
Speaker: Canan Gunes Corlu
Time: 12:15 PM - 1:30 PM
Location:
One of the challenges of developing digital twin simulations stems from lacking full information about business process flows and characterizations of their input distributions. This chapter describes how this challenge arises in different phases of digital twin development. We present practitioners an overview of solutions to use for correctly quantifying the overall uncertainty in digital twin simulation development. We accompany the presentation with a supply chain use case.
12:00 PM - 5:00 PM
TBD
Operations Spring Conference
-
Operations Spring Conference
Speaker: TBD
Time: 12:00 PM - 5:00 PM
Location:
8:00 AM - 1:00 PM
TBD
Operations Seminar Conference
-
Operations Seminar Conference
Speaker: TBD
Time: 8:00 AM - 1:00 PM
Location:
12:15 PM - 1:45 PM
Xiaoyun Qiu
Mechanism Design under Costly Signaling: the Value of Non-Coordination
-
Mechanism Design under Costly Signaling: the Value of Non-Coordination
Speaker: Xiaoyun Qiu
Time: 12:15 PM - 1:45 PM
Location:
We study optimal allocation mechanisms that rely on costly signals as screening devices. A social planner seeks to maximize welfare while ensuring the implementation of a given allocation rule. Allowing signal recommendations to depend on all agents’ reports (i.e., enabling information leakage) can improve coordination and lower the signaling costs for losers. However, this comes at the expense of inducing excessive effort from winners. We show that when higher types generate higher certainty-equivalent signals, the incentive cost of coordination outweighs its benefit. In this case, the optimal design is zero coordination, where signals are recommended solely based on each agent’s own report. We demonstrate that such non-coordination mechanisms can be implemented through coarse-ranking contests.
12:15 PM - 1:45 PM
Soroush Saghafian
Ambiguous Dynamic Treatment Regimes: A Reinforcement Learning Approach
-
Ambiguous Dynamic Treatment Regimes: A Reinforcement Learning Approach
Speaker: Soroush Saghafian
Time: 12:15 PM - 1:45 PM
Location:
A main research goal in various studies is to use an observational data set and provide a new set of counterfactual guidelines that can yield causal improvements. Dynamic Treatment Regimes (DTRs) are widely studied to formalize this process and enable researchers to find guidelines that are both personalized and dynamic. However, available methods in finding optimal DTRs often rely on assumptions that are violated in real-world applications (e.g., medical decision making or public policy), especially when (a) the existence of unobserved confounders cannot be ignored, and (b) the unobserved confounders are time varying (e.g., affected by previous actions). When such assumptions are violated, one often faces ambiguity regarding the underlying causal model that is needed to be assumed to obtain an optimal DTR. This ambiguity is inevitable because the dynamics of unobserved confounders and their causal impact on the observed part of the data cannot be understood from the observed data. Motivated by a case study of finding superior treatment regimes for patients who underwent transplantation in our partner hospital (Mayo Clinic) and faced a medical condition known as new-onset diabetes after transplantation, we extend DTRs to a new class termed Ambiguous Dynamic Treatment Regimes (ADTRs), in which the causal impact of treatment regimes is evaluated based on a “cloud” of potential causal models. We then connect ADTRs to Ambiguous Partially Observable Markov Decision Processes (APOMDPs) proposed by Saghafian (2018), and consider unobserved confounders as latent variables but with ambiguous dynamics and causal effects on observed variables. Using this connection, we develop two reinforcement learning methods termed Direct Augmented V-Learning (DAV-Learning) and Safe Augmented V-Learning (SAV-Learning), which enable using the observed data to effectively learn an optimal treatment regime. We establish theoretical results for these learning methods, including (weak) consistency and asymptotic normality. We further evaluate the performance of these learning methods both in our case study (using clinical data) and in simulation experiments (using synthetic data). We find promising results for our proposed approaches, showing that they perform well even compared with an imaginary oracle who knows both the true causal model (of the data-generating process) and the optimal regime under that model. Finally, we highlight that our approach enables a two-way personalization; obtained treatment regimes can be personalized based on both patients’ characteristics and physicians’ preferences.