4 Presentations

• 
  02:45 PM - 03:07 PM

  Optimizing Emergency Response: The MicroSWAP Framework for Nonprofit Resource Sharing and Collaboration

  • Ozge Aygul, presenter, Worcester Polytechnic Institute
  • Andrew Trapp, Worcester Polytechnic Institute

  While nonprofit organizations (NPOs) are key emergency responders providing essential resources and services, community-focused NPOs often face challenges in quickly securing resources and coordinating with other organizations. To address these challenges, we introduce a novel framework, designed for nonprofit organizations operating under a strict horizon to share resources during emergencies. Extending on the SWAP platform for resource sharing, the MicroSWAP auction mechanism allows nonprofits responding to emergencies to declare their needs and available resources. Upon reaching a certain market saturation, our novel framework activates and identifies optimal resource exchanges while assigning and scheduling pickup and delivery tasks. Our approach eliminates redundant trips by leveraging horizontal vehicle routing collaboration to optimize routes, improving service speed, ensuring efficient coordination among nonprofits, and increasing collective impact. We demonstrate the model's scalability and effectiveness through real-world inspired case studies, highlighting its potential in the nonprofit sector for emergency response.

• 
  03:07 PM - 03:29 PM

  Fair and Efficient Nonprofit Inventory Management under Display-Dependent Demand

  • Mahyar Eftekhar, Arizona State Univertisy
  • Yuan Shi, presenter, MIT
  • Yanchong Zheng, MIT

  More than 40 million people in the US rely on charitable food assistance, making food banks essential for improving food access. However, food bank operations face significant challenges, including limited warehouse space and unpredictable demand. Motivated by warehouse data from a regional US food bank, we develop a novel model for nonprofit inventory management where a planner aims to balance display fairness and throughput efficiency in the presence of heterogenous resource types and constrained capacity. The planner has only partial control over demand by adjusting inventory display levels. We characterize optimal display policies, identify key drivers of the fairness-efficiency trade-off, and propose flow interventions to improve both objectives. Numerical simulations using real warehouse data demonstrate that optimal flow interventions can improve biweekly display fairness by 10% on average during supply shortages and throughput efficiency by 15% during warehouse congestion. These findings provide actionable insights to optimize food bank operations.

• 
  03:29 PM - 03:51 PM

  Workforce Scheduling for Efficient and Equitable Access to Education

  • David Bergman, presenter, University of Connecticut
  • Carlos Cardonha, University of Connecticut
  • Mahyar Eftekhar, Arizona State University
  • Mohsen Emadikhiav, Florida Atlantic University

  Nonprofits supporting schools with extracurricular programs face unique challenges in efficiently assigning their skilled trainers. This research stems from our collaboration with New York Sun Works (NYSW), a nonprofit organization committed to delivering sustainability science and climate education to K-12 students across New York City. In this talk, we present optimization models and algorithms designed to address variants of the workforce scheduling problems encountered by NYSW.

• 
  03:51 PM - 04:13 PM

  Altruistic Bandit Learning For One-to-Many Matching Markets

  • Peash Ranjan Saha, presenter, School of Computing, Queen's University
  • Salimur Choudhury, School of Computing, Queen's University
  • Kai Salomaa, School of Computing, Queen's University

  Present-day two-sided matching applications consider stochastic environments where the agents learn preferences iteratively through explorations. A competition for an agent in the learning phase can be resolved if its preference is known. The multi-arm bandit learning model for the two-sided matching with known preferences for one side is appealing. However, the issue of objectionable cardinality due to agents’ incompatibility still exists. Some learning agents may only be compatible with a few highly competitive agents. We propose an altruistic bandit learning model for one-to-many matching where agents sacrifice choices to disburden competition from the highly desired agents. The model decreases the chances of agents with limited pairing options remaining unmatched, and the cardinality increases as a result. This model is suitable for humanitarian applications with a stability-cardinality trade-off (e.g.: homeless-shelter matching). Our experimental results indicate a noticeable increment of cardinality while the sacrificing agents’ loss is optimistically acceptable.