Group Fairness + Hedge
Group Fairness + Hedge:
- 2021:Best-Case Lower Bounds in Online Learning If each of finitely many experts approximately satisfies a certain notion of group fairness, then a clever use of the Hedge algorithm (running it separately on each group) also approximately satisfies the same notion of group fairness while still enjoying $O(\sqrt{T})$ regret. 每個experts都滿足一點group fairness, 那用樹籬算法也可以滿足group fairness.
- blum2018preserving 講了很多Fairness的故事
- bechavod2019equal 講了Equal Opportunity相關文獻. 與Hedge無關
1## Group Fairness + Hedge
2
3### 1.
4@article{guzman2021best,
5 title={Best-case lower bounds in online learning},
6 author={Guzm{\'a}n, Crist{\'o}bal and Mehta, Nishant and Mortazavi, Ali},
7 journal={Advances in Neural Information Processing Systems},
8 volume={34},
9 year={2021}
10}
11
12### 1.1
13@article{blum2018preserving,
14 title={On preserving non-discrimination when combining expert advice},
15 author={Blum, Avrim and Gunasekar, Suriya and Lykouris, Thodoris and Srebro, Nathan},
16 journal={arXiv preprint arXiv:1810.11829},
17 year={2018}
18}
19
20### 1.1.1.
21@article{bechavod2019equal,
22 title={Equal opportunity in online classification with partial feedback},
23 author={Bechavod, Yahav and Ligett, Katrina and Roth, Aaron and Waggoner, Bo and Wu, Zhiwei Steven},
24 journal={arXiv preprint arXiv:1902.02242},
25 year={2019}
26}
27
Fairness of Exposure in Stochastic Bandits
- 2021: Fairness of Exposure in Stochastic Bandits : 有Hedge法
- 2021: ONLINE FAIR REVENUE MAXIMIZING CAKE DIVISION WITH NON-CONTIGUOUS PIECES IN ADVERSARIAL BANDITS : 有Hedge法
- 2021: A Unified Approach to Fair Online Learning via Blackwell Approachability : Blackwell Approachability 給各種Fair online bandit.
1## Fair Exposure in Stochastic Bandits
2
3## 2.
4@article{wang2021fairness,
5 title={Fairness of Exposure in Stochastic Bandits},
6 author={Wang, Lequn and Bai, Yiwei and Sun, Wen and Joachims, Thorsten},
7 journal={arXiv preprint arXiv:2103.02735},
8 year={2021}
9}
10
11## 2.1
12@article{ghodsi2021online,
13 title={Online Fair Revenue Maximizing Cake Division with Non-Contiguous Pieces in Adversarial Bandits},
14 author={Ghodsi, Mohammad and Mirfakhar, Amirmahdi},
15 journal={arXiv preprint arXiv:2111.14387},
16 year={2021}
17}
18
19## 2.2
20@article{chzhen2021unified,
21 title={A Unified Approach to Fair Online Learning via Blackwell Approachability},
22 author={Chzhen, Evgenii and Giraud, Christophe and Stoltz, Gilles},
23 journal={Advances in Neural Information Processing Systems},
24 volume={34},
25 year={2021}
26}
27
樹籬算法
- 2022: Relaxing the I.I.D. Assumption: Adaptively Minimax Optimal Regret via Root-Entropic Regularization. (NIPS2022)
- 2019: On the optimality of the Hedge algorithm in the stochastic regime
- 2010: Prediction with expert advice under discounted loss. :Hedge算法的標準bound
鏡面下降
- Lecture note 19: Mirror descent ; Course
- 2019: Connections Between Mirror Descent, Thompson Sampling and the Information Ratio
評論