EMERGENT: Efficient and Manipulation-resistant Matching using GFlowNets
This addresses the problem of designing fair and efficient resource allocation algorithms for public services like housing or medical residency, offering a novel approach to balance efficiency and manipulability, though it is incremental in applying GFlowNets to a known bottleneck.
The paper tackled the trade-off between efficiency and strategyproofness in one-sided matching problems, such as school admissions, by proposing EMERGENT, a method using Generative Flow Networks (GFlowNets) that outperforms Random Serial Dictatorship in rank efficiency and reduces strategic vulnerability compared to Rank Minimization and Probabilistic Serial.
The design of fair and efficient algorithms for allocating public resources, such as school admissions, housing, or medical residency, has a profound social impact. In one-sided matching problems, where individuals are assigned to items based on ranked preferences, a fundamental trade-off exists between efficiency and strategyproofness. Existing algorithms like Random Serial Dictatorship (RSD), Probabilistic Serial (PS), and Rank Minimization (RM) capture only one side of this trade-off: RSD is strategyproof but inefficient, while PS and RM are efficient but incentivize manipulation. We propose EMERGENT, a novel application of Generative Flow Networks (GFlowNets) to one-sided matching, leveraging its ability to sample diverse, high-reward solutions. In our approach, efficient and manipulation-resistant matches emerge naturally: high-reward solutions yield efficient matches, while the stochasticity of GFlowNets-based outputs reduces incentives for manipulation. Experiments show that EMERGENT outperforms RSD in rank efficiency while significantly reducing strategic vulnerability compared to matches produced by RM and PS. Our work highlights the potential of GFlowNets for applications involving social choice mechanisms, where it is crucial to balance efficiency and manipulability.