November 2023 - Ml-reading-group - lists.saarland-informatics-campus.de

Our next reading group: Wednesday, Nov. 29 @ 14:30, Why does Throwing Away Data Improve Worst-Group Error?
by Huynh Quang Kiet Vo 23 Nov '23

23 Nov '23

Dear all, Our next reading group is scheduled as follows: Paper: Why does Throwing Away Data Improve Worst-Group Error? [1] Presenter: Jake Fawkes [2] Date/Time: Wednesday 29.11.2023, 14:30 - 16:00 Location: CISPA C0 Building: 0.01 Abstract: When facing data with imbalanced classes or groups, practitioners follow an intriguing strategy to achieve best results. They throw away examples until the classes or groups are balanced in size, and then perform empirical risk minimization on the reduced training set. This opposes common wisdom in learning theory, where the expected error is supposed to decrease as the dataset grows in size. In this work, we leverage extreme value theory to address this apparent contradiction. Our results show that the tails of the data distribution play an important role in determining the worst-group-accuracy of linear classifiers. When learning on data with heavy tails, throwing away data restores the geometric symmetry of the resulting classifier, and therefore improves its worst-group generalization. For more details about future arrangements, please see this sheet [here [3]]. Best Regards, Kiet Vo Links: ------ [1] https://arxiv.org/abs/2205.11672 [2] https://csml.stats.ox.ac.uk/people/fawkes/ [3] https://docs.google.com/spreadsheets/d/1vtgEezBqS4d_ACPt-emK2NT52x7nofX9jxg…

1 0

Invited talk by Kirtan Padh today!
by Adrián Javaloy 15 Nov '23

15 Nov '23

Dear all, This is a reminder that we will have Kirtan Padh <https://kirtan.netlify.app/> at the ML reading group today at 2:30pm. Please remember that this will be a *fully online session* (see details below). Best regards, Adrián On 10. Nov 2023, at 10:06, Adrián Javaloy <adrian.javaloy(a)cs.uni-saarland.de> wrote: Dear all, Next week we will have Kirtan Padh <https://kirtan.netlify.app/> as an invited speaker for the reading group. Kirtan is a PhD candidate at Helmholtz AI in Munich, advised by Prof. Niki Kilbertus (which some of us know from the MPI-IS times), and broadly speaking his work lays on the intersection of machine learning, causality, and fairness. This was a talk scheduled quite a bit in advance by @Miriam, which will be our host next week. Unfortunately, Kirtan is not able to come in person and, therefore, the reading group *will be online next week.* You only need to access the _Zoom room located at the top of the shared spreadsheet_. Please let me know if there is any doubts, and I hope to virtually see you all next week! * *Speaker:* Kirtan Padh * *Title of the talk:* Addressing Fairness in Classification with a Model-Agnostic Multi-Objective Algorithm * *Link to the related paper:* https://proceedings.mlr.press/v161/padh21a/padh21a.pdf <https://proceedings.mlr.press/v161/padh21a/padh21a.pdf?> * *Abstract:* The goal of fairness in classification is to learn a classifier that does not discriminate against groups of individuals based on sensitive attributes, such as race and gender. One approach to designing fair algorithms is to use relaxations of fairness notions as regularization terms or in a constrained optimization problem. We observe that the hyperbolic tangent function can approximate the indicator function. We leverage this property to define a differentiable relaxation that approximates fairness notions provably better than existing relaxations. In addition, we propose a model-agnostic multi-objective architecture that can simultaneously optimize for multiple fairness notions and multiple sensitive attributes and supports all statistical parity-based notions of fairness. We use our relaxation with the multi-objective architecture to learn fair classifiers. Experiments on public datasets show that our method suffers a significantly lower loss of accuracy than current debiasing algorithms relative to the unconstrained model. Cheers, Adrián

1 0

Invited talk by Kirtan Padh next week!
by Adrián Javaloy 10 Nov '23

10 Nov '23

Dear all, Next week we will have Kirtan Padh <https://kirtan.netlify.app/> as an invited speaker for the reading group. Kirtan is a PhD candidate at Helmholtz AI in Munich, advised by Prof. Niki Kilbertus (which some of us know from the MPI-IS times), and broadly speaking his work lays on the intersection of machine learning, causality, and fairness. This was a talk scheduled quite a bit in advance by @Miriam, which will be our host next week. Unfortunately, Kirtan is not able to come in person and, therefore, the reading group *will be online next week.* You only need to access the _Zoom room located at the top of the shared spreadsheet_. Please let me know if there is any doubts, and I hope to virtually see you all next week! * *Speaker:* Kirtan Padh * *Title of the talk:* Addressing Fairness in Classification with a Model-Agnostic Multi-Objective Algorithm * *Link to the related paper:* https://proceedings.mlr.press/v161/padh21a/padh21a.pdf <https://proceedings.mlr.press/v161/padh21a/padh21a.pdf?> * *Abstract:* The goal of fairness in classification is to learn a classifier that does not discriminate against groups of individuals based on sensitive attributes, such as race and gender. One approach to designing fair algorithms is to use relaxations of fairness notions as regularization terms or in a constrained optimization problem. We observe that the hyperbolic tangent function can approximate the indicator function. We leverage this property to define a differentiable relaxation that approximates fairness notions provably better than existing relaxations. In addition, we propose a model-agnostic multi-objective architecture that can simultaneously optimize for multiple fairness notions and multiple sensitive attributes and supports all statistical parity-based notions of fairness. We use our relaxation with the multi-objective architecture to learn fair classifiers. Experiments on public datasets show that our method suffers a significantly lower loss of accuracy than current debiasing algorithms relative to the unconstrained model. Cheers, Adrián

1 0

2024

2023

Ml-reading-group November 2023

2024

2023

Ml-reading-group November 2023 ----- 2024 ----- May 2024 April 2024 March 2024 February 2024 January 2024 ----- 2023 ----- December 2023 November 2023 October 2023

Ml-reading-group November 2023