News

I feel very honored to have received the Emerging Alumni Leader Award from Imperial College London, where I did both my undergrad, earning an M.Eng. in Electrical & Electronic Engineering, and my PhD in Computing. See here for more information the award.

Our paper on "Learning fair policies for multi-stage selection problems from observational data" has been accepted for publication at AAAI! 

In this paper, we consider the problem of learning fair policies for multi-stage selection problems from observational data. This problem arises in several high-stakes domains such as company hiring, loan approval, or bail decisions where outcomes (e.g., career success, loan repayment, recidivism) are only observed for those selected. We propose a multi-stage framework that can be augmented with various fairness constraints, such as demographic parity or equal opportunity. This problem is a highly intractable infinite chance-constrained program involving the unknown joint distribution of covariates and outcomes.  Motivated by the potential impact of selection decisions on people’s lives and livelihoods, we propose to focus on interpretable linear selection rules. Leveraging tools from causal inference and sample average approximation, we obtain an asymptotically consistent solution to this selection problem by solving a mixed binary conic optimization problem, which can be solved using standard off-the-shelf solvers. We conduct extensive computational experiments on a variety of datasets adapted from the UCI repository on which we show that our proposed approaches can achieve an 11.6\% improvement in precision and a 38\% reduction in the measure of unfairness compared to the existing selection policy.  

Check-out this article about these news and about my vision for the Center: https://viterbischool.usc.edu/news/2023/08/phebe-vayanos-named-co-director-of-usc-center-for-artificial-intelligence-in-society/

Register for the TED AI event here: https://www.ai-event.ted.com 

In this paper, we study systems that allocate scarce resources to satisfy basic needs, such as homeless services that provide housing. We propose a framework for evaluating fairness in contextual resource allocation systems that is inspired by fairness metrics in machine learning. This framework can be applied to evaluate the fairness properties of a historical policy, as well as to impose constraints in the design of new (counterfactual) allocation policies. 

Our work culminates with a set of incompatibility results that investigate the interplay between the different fairness metrics we propose. Notably, we demonstrate that: 1) fairness in allocation and fairness in outcomes are usually incompatible; 2) policies that prioritize based on a vulnerability score will usually result in unequal outcomes across groups, even if the score is perfectly calibrated; 3) policies using contextual information beyond what is needed to characterize baseline risk and treatment effects can be fairer in their outcomes than those using just baseline risk and treatment effects; and 4) policies using group status in addition to baseline risk and treatment effects are as fair as possible given all available information. Our framework can help guide the discussion among stakeholders in deciding which fairness metrics to impose when allocating scarce resources.

See here for more details: https://arxiv.org/abs/2212.01725

Check it out: "Learning resource allocation policies from observational data with an appllication to homeless services delivery"!

I am thrilled to annouce that Nathan Justin, second year PhD student in our group in Computer Science, and Bill Tang, second year PhD student in our group in ISE, have both  been awarded the prestigious Graduate Research Fellowship Program (GRFP) award from the National Science Foundation. Check-out this article about their award and research.

Check out our new paper on "Learning optimal fair classification trees"!

Check out our new paper on "Learning resource allocation policies from observational data with an appllication to homeless services delivery"!

Our paper on "Learning optimal prescriptive trees from observational data" has been accepted at the AAAI Workshop on AI for Behavior Change!

Our paper on "Optimal robust classification trees" has been accepted at the AAAI Workshop on Adversarial Machine Learning and Beyond!

I'll be teaching a new course on "Analytics for Social Impact" this coming Spring!

This course will introduce students to research in the emerging area of “Analytics for Social Impact” whose goal is to advance descriptive, predictive, and prescriptive analytics to address important societal challenges, e.g., in public health, social services provision, and conservation. Particular methods that the students will learn about relate to fairness, interpretability, and robustness, from the lens of optimization, machine learning, and causal inference.

E-mail me for more information.

Our new paper on "Learning optimal prescriptive trees from observational data" in now online. In this work, we consider the problem of learning an optimal prescriptive tree (i.e., a personalized treatment assignment policy in the form of a binary tree) of moderate depth, from observational data. This problem arises in numerous socially important domains such as public health and personalized medicine, where interpretable and data-driven interventions are sought based on data gathered in deployment, through passive collection of data, rather than from randomized trials.

I am honored to have been appointed as holder of a WiSE Gabilan Assistant Professorship.

An updated version of our paper on "Robust Optimization with Decision-Dependent Information Discovery" is now available online. It includes applications of our approach to R&D project porfolio selection and to the Pandora Box problem. We also evaluate our method on an instance of the active preference elicitation problem used to recommend kidney allocation policies to policymakers at the United Network for Organ Sharing based on real data from the U.S. Kidney Allocation System.

Our new paper on "Strong Optimal Classification Trees" is now under review at Operations Research. Our proposed approach is 31 times faster than state-of-the art MIO-based techniques and improves out of sample performance by up to 8%. Our formulation can accommodate side constraints to enable the design of interpretable and fair decision trees.

I am really excited to announce that, starting April 2021, I will be appointed Associate Editor for Operations Research Letters in the Optimization under Uncertainty and Machine Learning area.

Our paper on "Fair Influence Maximization: a Welfare Optimization Approach" has been accepted for publication in Proceedings of the AAAI Conference on Artificial Intelligence, 2020.

Organized explOR, a day-long program aimed to bring STEM education to underserved communities, focusing on topics in Operations Research (OR) and Artificial Intelligence (AI) applied to problems of Social Good in collaboration with educators, staff, and students at Code in the Schools and STEM Academy of Hollywood

Our paper on "Cost-Sharing Mechanism Design for Ride-Sharing" has been submitted for publication.

We are very excited to have received a USC Zumberge Faculty Research and Innovation Fund Award to support our project on "Addressing Ethical Dilemmas during the COVID-19 Pandemic through AI."

Our paper "ROC++: Robust Optimization in C++" has been submitted for publication.

One of the major problems that has emerged during the COVID-19 pandemic is that hospitals and states have more cases than they have resources to respond to and treat positive patients (e.g., not enough personal protective equipment [PPE], not enough ventilators, not enough doctors, not enough nurses). This has placed decision makers in these systems (e.g., doctors and policymakers) in the extremely difficult position to have to choose who gets access to and priority for these resources. This poses huge ethical dilemmas (e.g., need to balance system efficiency with fairness considerations) which are complicated by the diversity of stakeholders in these systems (e.g., doctors, patients, hospitals, governments, population at-large) and by the high degree of uncertainty under which decisions must be made. Thus, there is an urgent need for disciplined, automated, data-driven approaches for coordinating the allocation of these scarce resources.

Our paper on "Active Preference Elicitation via Adjustable Robust Optimization" with an application to learn the preferences of policy-makers to be able to design policies that better meet their needs has been submitted for publication.

Our paper on "Learning Optimal Classification Trees: Strong Max-Flow Formulations" is now under review for publication in the proceedings of ICML.

Our paper on "Robust Optimization with Decision-Dependent Information Discovery" with an application to preference elicitation at the Los Angeles Homeless Services Authority has been submitted for publication.

We are grateful that our paper on "Exploring Algorithmic Fairness in Robust Graph Covering Problems" has been accepted for publication in the Proceedings of NeurIPS 2019. This paper proposes an algorithm for designing fair "Gatekeeper" training interventions for suicide prevention. We hope our work can help raise awareness of this important problem.

We are excited that our work with my colleague Eric Rice in collaboration with the Los Angeles Homeless Services Authority (LAHSA) is featured in One Zero Medium

Together with Behavioral/Social Scientists, Physical Scientists, and Economists at RAND Corporation, Geological and Environmental Scientists at the University of Oregon, and partners at the Sitka Sound Science Center, we have been awarded an NSF Smart & Connected Cities $2,100,974 grant for our project "Landslide Risk Management in Remote Communities: Integrating Geoscience, Data Science, and Social Science in Local Context."

Together with my colleague, Bistra Dilkina, we are really excited to have been awarded a $535,335 NSF grant for our project "Preserving Biodiversity via Robust Optimization."