| 研究課題/領域番号 |
20K11947
|
|---|
| 研究機関 | 国立研究開発法人産業技術総合研究所 |
|---|
研究代表者 |
Nicolas Schwind 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究員 (60646397)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2023-03-31
|
| キーワード | Resilience / Multiagent systems / Facility Location / Team Formation |
|---|
| 研究実績の概要 |
Our main contribution for FY2020 is manifold: (i) we introduced a novel notion of resilience called partial robustness in multiagent systems and team formation; (ii) we analyzed the computational complexity of forming a partially robust team, which turns out to be in-between the previous existing notions of robustness and recoverability in team formation; (iii) we provided a complete algorithm for computing partially robust teams; (iv) we implemented our algorithm and compared its performance with the existing previous notions of robustness and recoverability on a number of existing and new benchmarks. Our findings were that in practice, our new notion is shown to be an interesting trade-off notion between (full) robustness and recoverability in terms of initial team deployment cost, efficiency coverage in the disaster phase of a resilience scenario, and recovery cost in the repair phase. In addition, our new notion is much more efficient than previous resilience ones, and thus more suited to be used for reasonably sized problem instances. Lastly, we have provided another probabilistic notion of resilience in multiagent systems in the context of coalition structure generation.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
1: 当初の計画以上に進展している
理由
Our new resilience notion was initially based on the intuition that it would be an interesting trade-off between existing ones, in terms of computational efficiency and solution quality. Both criteria turned out to be fulfilled above our initial expectations: (i) computational efficiency: about 90% of realistic large-scale instances could be solved while a previous resilience notion introduced in 2018 could not solved any of them; (ii) while guaranteeing near-optimal quality deployments in terms of resilience, the cost of deployment can be shown to be reduced by 50% in some cases. As a result, our main paper for FY2020 has been published to AAMAS 2021, the premier international conference on multiagent systems. In addition, our paper has been recognized as a premier paper of the conference series and has been recommended to be extended and submitted to the journal of AAMAS in a fast-publication track. Lastly, our instance generator has been released publicly (https://github.com/jm62300/team-formation) as well as our algorithms, and so is already available for public use.
|
| 今後の研究の推進方策 |
In perspective, we plan to pursue the development of our publicly released instance generator and algorithms to make it more user-friendly. The development will include the tuning of some parameters of Perlin algorithm used to generated realistic instances of various kinds. Another perspective for benchmark generation is, as initially planned in the project proposal, to take advantage of real-world populated maps and translate them into our facility location problem tam formation instances. In addition, we will pursue our theoretical works on resilience in terms of the system's accommodation to change.
|
| 次年度使用額が生じた理由 |
The initial plan was to use the fundings for travelling purpose to international conferences and to visit the collaborators of this project in CRIL, University of Artois, France. The coronavirus situation made this plan impossible, so only a small amount of the budget was used to subscribe to an online conference (KR 2020). The budget will be used as soon as travelling is made possible again, which will help making boosting research work by physically meeting with the project collaborators.
|
