Building Sequential Resource Allocation Mechanisms without Payments
Keywords: Sequential resource allocation, mechanism design, incentive compatibility
Abstract: We study allocating divisible resources of limited quantities to agents who submit requests for the resources one or multiple times over a finite horizon. This is referred to as the sequential or online resource allocation problem, as irrevocable allocations need to be made as the requests arrive, without observations on the future requests. The existing work on sequential resource allocation (in the payment-free setting) mainly focuses on optimizing social welfare and designs mechanisms under the assumption that the agents make truthful requests. Such mechanisms can be easily exploitable -- strategic agents may misreport their requests to inflate their allocations. Our aim in this work is to design sequential resource allocation mechanisms that balance the competing objectives of social welfare maximization (promoting the overall agent satisfaction) and incentive compatibility (ensuring that the agents do not have incentives to misreport). We do not design these mechanisms from scratch. Instead, as the incentive compatible mechanism design problem has been well studied in the one-shot setting (horizon length equals one), we propose a general meta-algorithm of transforming a one-shot mechanism into its sequential counterpart. The meta-algorithm can plug in any one-shot mechanism and approximately carry over the properties that the one-shot mechanism already satisfies to the sequential setting. We establish theoretical results validating these claims and illustrate their superior performance relative to baselines in experiments.
Submission Number: 251
Loading