Multi-Unit Auctions for Allocating Chance-Constrained Resources
Authors: Anna Gautier, Bruno Lacerda, Nick Hawes, Michael Wooldridge
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we demonstrate empirically that our auction outperforms state-of-the-art techniques for chance-constrained multi-agent resource allocation in complex settings with up to hundreds of agents. To evaluate the performance of ACCR, we compared its performance to a series of baselines on the benchmark domain Maze (Wu and Durfee 2010), and an advertising budget allocation MDP from Boutilier and Lu (2016). |
| Researcher Affiliation | Academia | Anna Gautier, Bruno Lacerda, Nick Hawes, Michael Wooldridge University of Oxford Oxford, United Kingdom anna.gautier@eng.ox.ac.uk, bruno@robots.ox.ac.uk, nickh@robots.ox.ac.uk, mjw@cs.ox.ac.uk |
| Pseudocode | No | The paper describes procedures like 'The ACCR Protocol' and 'Single-Agent Decision Making' in paragraph format, but it does not include a formally structured 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper does not provide any concrete access information for source code, such as a repository link or an explicit statement about code release in supplementary materials. |
| Open Datasets | Yes | To evaluate the performance of ACCR, we compared its performance to a series of baselines on the benchmark domain Maze (Wu and Durfee 2010), and an advertising budget allocation MDP from Boutilier and Lu (2016). |
| Dataset Splits | No | The paper describes the domains and the number of trials for evaluation (e.g., 'average over 50 trials'), but does not specify explicit training, validation, and test dataset splits or percentages. |
| Hardware Specification | Yes | All experiments were conducted on an AWS R5a.large EC2 instance, with 2 CPUs and 16GB of memory. |
| Software Dependencies | No | For all methods, LPs and MILPs were implemented with Gurobi, and all MDP methods (e.g., solving the MMDPs, computing maximum reward policies for CG, computing Pareto frontiers) were solved using the PRISM model checker (Kwiatkowska, Norman, and Parker 2002). The software components (Gurobi, PRISM) are mentioned, but specific version numbers are not provided. |
| Experiment Setup | Yes | In all experiments, we bound the probability of resource violations with a chance constraint of δ = 0.05. The global resource constraint is set to L = hn/4, where h is the global time horizon and n is the number of agents in the system. As in Boutilier and Lu (2016) the time horizon is 50, though we modify the objective to undiscounted reward. For our ACCR algorithm, agents restrict their bids by only generating Pareto frontiers for k divisible by 10. |