This theme considers problems where there are multiple objectives we care about – these might come from various stakeholder preferences, notions of fairness, statistical metrics in machine learning settings, regret guarantees, computational efficiency, variance in decisions, etc.

(a) Portfolios: The theory of multi-criteria optimization to simultaneously optimize for efficiency and fairness is not new. However, it is often unclear how to justify selection of a specific notion of fairness to achieve balance or efficiency in resource allocation. A resource allocation problem driven by efficiency can over-allocate to certain demographics, while neglecting other groups. To explore this idea, we introduced the notion of solution portfolios: given a class C of fair objectives, find a small set of solutions S so that for any given function f(.) in C, there exists a solution s in S which is a good approximation for optimal under function f(.).

For various combinatorial problems, we are interested in mathematically quantifying the trade-off between the size of the portfolio, and approximation factors possible for various classes of balancing objectives C. Constructing these portfolios gives important actionable insight to decision-makers, as they can bypass the choice of the model, but instead focus on the properties of the solutions in the portfolio. Our work on facility location attempts to answer the policy question on developing infrastructure to reduce the number of medical deserts in the US (e.g., explore our app here). We extended this concept to the space of reinforcement learning policies, which allows us to navigate stakeholder preferences in episodic RL, human-AI collaboration through transparent dashboards, and interface with LLMs in policy design.

  • Navigating the Social Welfare Frontier: Portfolios for Multi-objective Reinforcement Learning, Cheol Woo Kim, Jai Moondra, Shresth Verma, Madeleine Pollack, Lingkai Kong, Milind Tambe, Swati Gupta, ICML 2025. arxiv
  • Approximation Algorithms for Fair Portfolio of Solutions in Combinatorial Optimization – Swati Gupta, Jai Moondra, Mohit Singhh, SODA 2025, arxiv.
  • Which Lp norm is the fairest? Approximations for fair facility location across all “p” – Swati Gupta, Jai Moondra, Mohit Singh. EC 2023, arxiv.
  • Too many fairness metrics: Is there a solution? – Swati Gupta, Akhil Jalan, Gireeja Ranade, Helen Yang, Simon Zhuang, Fields Institute Communication Series (forthcoming) and EDSC 2020

(b) Trajectory-Constrained Online Learning: We are interested in developing the theory of online decision-making and iterative optimization, by incorporating trajectory-constraints on the iterates of the algorithm. This helps bring in the perspective of fairness of decisions that are taken over a period of time. These decisions often interact with people, who feel mistreated when they receive decisions that change over time. We considered how a popular notion of individual fairness – that similar individuals should be treated similarly – and proposed an extension over time that generalizes concepts like stare-decisis (in law), markdowns (in retail), and whataboutisms (in politics).

In the context of demand learning for multiple segments of customers, with inter-segment constraints such as price offered to students should be at most the price offered to the general population, we showed that there is no gap in achievable regret rates with and without monotonicity constraints, for smooth and strongly convex functions, while respecting a certain class of inter-segment constraints. These questions are important also in the context of platforms like Upwork, where there is a need to balance opportunities while ensuring that the platform thrives. We explored new algorithmic techniques for allocating jobs, given different learning and performance over different gig economy workers.

  • Algorithmic Challenges in Ensuring Fairness at the Time of Decision with Salem and Kamble, WINE 2022, and Operations Research 2025.
  • Temporal Fairness in Online Decision-Making, Swati Gupta, Jad Salem, Vijay Kamble, Book Chapter in Ethics in Artificial Intelligence: Bias, Fairness and Beyond, 2023.
  • Individual Fairness in Hindsight, Swati Gupta, Vijay Kamble, Journal of Machine Learning Research 2021.
  • Group-Fair Online Allocation in Continuous Time with Semih Cayci, Swati Gupta, Atilla Eryilmaz, NeurIPS 2020.

(c) Districting and Gerrymandering: To understand fair resource allocation in the context of public safety and representation, we looked into districting problems: finding balanced and continuous partitions of a given city so that the parts of the partition are compact. There exist many graph partitioning methods previously, however, these do not guarantee contiguity. In fact, even existing heuristics for continuous graph partitioning were not able to compute a good districting for balancing 911 call workloads for the South Fulton City due to size of the instances. We explored multi-criteria combinatorial approximation methods to find good districting plans for unweighted grid graphs, and heuristic extensions of these to (stochastic) weighted instances. This allowed our algorithms to scale to large instances, and one of the plans from this work was implemented in the SF City. We also looked into gerrymandering in Georgia’s Congressional and state districting plans, and showed that the current plan of Georgia is largely unresponsive to the changing opinions of the electorate (i.e., there is no point of holding elections in GA).

  • Balanced Districting on Grid Graphs with Provable Compactness and Contiguity –
    Cyrus Hettle, Shixiang Zhu, Swati Gupta, Yao Xie, FORC 2021
  • Mathematically Quantifying Non-responsiveness of the 2021 Georgia Congressional Districting Plan – Zhanzhan Zhao, Cyrus Hettle, Swati Gupta, Jonathan C. Mattingly, Dana Randall, Gregory J. Herschlag, EAAMO 2022

(d) Navigating ML and OR Pipelines: We have been exploring various multi-criteria decision-making scenarios in supply chains and power systems. There are many interesting connections between data privacy, algorithmic fairness and reliability in networks under environmental disruptions such as wildfires near transmission networks, and other local grid failures. The overarching question about the interactions of societal operations research with the census data, where OR systems interact with noisy data and ML pipelines provide a massive opportunity for AI and urgent need for reliability and resilience. We are exploring this from various perspectives of robust, stochastic optimization, multi-criteria optimization and policy making.

  • Equitably allocating wildfire resilience investments for power grids: The curse of aggregation and vulnerability indices – Madeleine Pollack, Ryan Piansky, Swati Gupta and Daniel Molzahn, Applied Energy, 2025
  • Fair and Reliable Reconnections for Temporary Disruptions in Electric Distribution Networks – Swati Gupta, Cyrus Hettle and Daniel Molzahn, INFORMS Journal on Computing 2025
  • Generating clusters for urban logistics in hyperconnected networks with Hettle, Faugere, Kwon, Montreuil International Physical Internet Conference IPIC 2021
  • Robust Look-ahead Three-phase Balancing of Uncertain Distribution Loads – Xinbo Geng, Swati Gupta, Le Xie, HICCS 2019
  • Fairness in Inventory Routing with Michael Wang, NeurIPS Workshop on Ethical, Social and Governance Issues in AI, 2018