[論文レビュー] A Comprehensive Survey of Incentive Mechanism for Federated Learning

Rongfei Zeng, Chao Zeng+3 more|arXiv (Cornell University)|Jun 27, 2021

Privacy-Preserving Technologies in Data参考文献 52被引用数 59

ひとこと要約

連邦学習のインセンティブ機構の系統的調査で、問題定式化、分類、主要技術（Shapley value、Stackelberg games、auctions、contracts、RL、blockchain）、および今後の方向性を詳述する。

ABSTRACT

Federated learning utilizes various resources provided by participants to collaboratively train a global model, which potentially address the data privacy issue of machine learning. In such promising paradigm, the performance will be deteriorated without sufficient training data and other resources in the learning process. Thus, it is quite crucial to inspire more participants to contribute their valuable resources with some payments for federated learning. In this paper, we present a comprehensive survey of incentive schemes for federate learning. Specifically, we identify the incentive problem in federated learning and then provide a taxonomy for various schemes. Subsequently, we summarize the existing incentive mechanisms in terms of the main techniques, such as Stackelberg game, auction, contract theory, Shapley value, reinforcement learning, blockchain. By reviewing and comparing some impressive results, we figure out three directions for the future study.

研究の動機と目的

• Identify the incentive problem in federated learning and its goal to improve FL performance.
• Provide a comprehensive taxonomy of incentive schemes in FL across settings, phases, and techniques.
• Summarize existing mechanisms by main techniques and sub-problems, with emphasis on contribution evaluation, node selection, and payments.
• Highlight assumptions, advantages, and limitations to guide future research in FL incentives.

提案手法

• Define the incentive mechanism for FL with multi-dimensional contributions and model owner payments.
• Classify incentive schemes by application setting, FL phase, main techniques, sub-problems, and information symmetry.
• Review mechanisms using Shapley value, Stackelberg games, auctions, contract theory, reinforcement learning, blockchain, and other approaches.
• Discuss properties like IC, IR, fairness, PE, CR, BB, and introduce Performance Improvement (PI) as key for FL.
• Provide examples, discuss computational/privacy challenges, and compare representative works.

実験結果

リサーチクエスチョン

• RQ1 What constitutes an effective incentive mechanism in Federated Learning and how does it impact FL performance?
• RQ2 How are incentive schemes in FL constructed across cross-device and cross-silo settings, and which techniques are most effective in different sub-problems (contribution evaluation, node selection, payment allocation)?
• RQ3 What assumptions (information symmetry) do existing schemes rely on, and what are their implications for practicality and robustness?
• RQ4 What are the key future directions and open challenges in designing FL incentive mechanisms?

主な発見

• Shapley value は寄与度評価に広く用いられるが計算コストが高く、さまざまな近似法が提案されている。
• Stackelberg games および auctions は、FL におけるリーダー-フォロワーの支払いとリソース割り当て問題で一般的に用いられている。
• Contract theory、 reputation systems、 blockchain は、情報の非対称性、信頼性、頑健性をFLのインセンティブに対処するために用いられている。
• Reinforcement learning は、情報が不完全な状況下での動的戦略適応の機構を提供する。
• Prototype schemes は、適切に設計されたインセンティブを用いるとFLの訓練時間とモデル精度の性能改善を示す。
• 将来の研究はトレーニング性能、MEC/5G/IoT 制約、クロスシロ FL のインセンティブに焦点を当てるべきである。