【联邦学习】激励机制

总览

横向联邦学习中的数据标签质量差异问题

请专家在服务器端做一个质量比较高的数据集。

Y. Chen, X. Yang, X. Qin, H. Yu, B. Chen & z. Shen, "FOCUS: Dealing with Label Quality Disparity in Federated Learning,'CoRR, arXiv:2001.11359, 2020.

纵向联邦学习中的参与方和特征重要性评估问题

Siwei Feng & Han Yu, “Multi-Participant Multi-Class Vertical Federated Learning,” CoRR, arXiv:2001.11154, 2020.

如何通过奖惩措施应对及遏制联邦学习恶意参与方的问题

• Can we find a dynamic and cost-effective best response function for the defender that maximizes the attacker’s probability of selection“Do not Attack"?
• Defenders:
  – Faces many types of threats
  – Has a limited budget for screening submitted model parameters (different screening methods incur different costs)
  –Can announce punitive measures before-hand

Following the Stackelberg Game formulation, since the leader (attacker) will make the first move, she knows that a rational follower (the data federation) will react by maximizing the follower’s payoff. The attacker takes this into account before making the first move.

L. Lyu, H. Yu & Q. Yang, “Threats to Federated Learning: A Survey,” CoRR, arXiv:2003.02133, 2020.

如何去除参与顺序对贡献度的影响的问题

A. Ghorbani & J. Zou, " Data Shapley: Equitable Valuation of Data for Machine Learning," CoRR, arXiv:1904.02868, 2019.

1. More powerful than the popular leave one-out or leverage score in providing insight on what data is
   more valuable for a given learning task
2. Low Shapley value data effectively capture outliers and corruptions
3. High Shapley value data inform what type of new data to acquire to improve the predictor.

挑战
Data Shapley值计算复杂度非常高，不适合由联邦学习参与方直接计算。

Y. Liu, S. Sun, Z. Ai, S. Zhang, z. Liu & H. Yu, “FedCoin: A Peer-to-Peer Payment System for Federated Learning,” CoRR,arXiv:2002.11711, 2020.

如何设计公平、可解释的联邦学习利益分配方案的问题

Yu, H., Liu, Z., Liu, Y., Chen, T., Cong, M., Weng, X., Niyato, D. & Yang, Q. A sustainable incentive scheme for federatedlearning. IEEE Intelligent Systems PP(99), IEEE (2020).
H. Yu, z. Liu, Y. Liu, T. Chen, M. Cong, X. Weng, D. Niyato & Q. Yang, “A Fairness-aware Incentive Scheme for Federated Learning,” in Proceedings of the 3rd AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society (AIES- 20), pp. 393-399, 2020.

如何研究参与方在不同激励机制下的反应的问题

• We propose a multi-player game to study how FL participants make action selection decisions under
  different incentive schemes.
• It allows human players to role-play under various conditions to guide FL incentive research.

视频简介: https://youtu. be/4qd48QfcsXI

K.L. Ng, z. Chen, z. Liu, H. Yu, Y. Liu & Q. Yang, “A Multi-player Game for Studying Federated Learning Incentive Schemes,” IJCAI, 2020 (under review).

联邦学习激励机制研究体系的愿景