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

[Cancelled RG] Worker's day --> Workers don't work
by Adrián Javaloy 15 May '24

15 May '24

Dear all, As you all are aware, this Wednesday is Worker's day and therefore a public holiday in Saarland. Therefore, we will *cancel this week's reading group* and will *resume it in two weeks* with Krikamol as speaker, talking about Credal Networks. I hope you have a good mid-week rest, and that you use the occasion to *not work* *at all*. Best regards, Adrián

3 2

Not So Fair: The Impact of Presumably Fair Machine Learning Models
by Adrián Javaloy 16 Apr '24

16 Apr '24

Dear ML reading group, This week we will have Deborah presenting the paper *Not So Fair: The Impact of Presumably Fair Machine Learning Models* (paper here <https://dl.acm.org/doi/10.1145/3600211.3604699>). The reading group will take place on *Wednesday at 2:30pm *as usual, and this time we will host physically at *room 0.01 of building E1.7 @ UdS*. For those attending remotely, you can find the Zoom link in the reading group spreadsheet <https://docs.google.com/spreadsheets/d/1vtgEezBqS4d_ACPt-emK2NT52x7nofX9jxg…>. *Abstract:* When bias mitigation methods are applied to make fairer machine learning models in fairness-related classification settings, there is an assumption that the disadvantaged group should be better off than if no mitigation method was applied. However, this is a potentially dangerous assumption because a “fair” model outcome does not automatically imply a positive impact for a disadvantaged individual—they could still be negatively impacted. Modeling and accounting for those impacts is key to ensure that mitigated models are not unintentionally harming individuals; we investigate if mitigated models can still negatively impact disadvantaged individuals and what conditions affect those impacts in a loan repayment example. Our results show that most mitigated models negatively impact disadvantaged group members in comparison to the unmitigated models. The domain-dependent impacts of model outcomes should help drive future bias mitigation method development. Cheers, Adrián

2 1

Wednesday Reading Group
by Siu Lun Chau 20 Mar '24

20 Mar '24

Dear all, This week’s reading group is hosted by CISPA, and we will have Anurag Singh presenting the paper on "Invariant Risk Minimisation Game" (https://arxiv.org/abs/2002.04692). The reading group will be on Wednesday at 2:30 pm. As you may know, CISPA has recently moved offices to St.Ingbert so the organisers have proposed the following solution to keep the reading group running: Both udS and CISPA book a room for their members to meet and the presenter will present online in their respective location. The link to join virtually can be found here: https://docs.google.com/spreadsheets/d/1vtgEezBqS4d_ACPt-emK2NT52x7nofX9jxg… Thank you, Alan

2 2

Updated paper for today's RG
by Pablo 06 Mar '24

06 Mar '24

Hello All, I've decided to change the paper for today's reading group. The proposed one had too much text and it would not be an "enjoyable” experience for the RG! This is the paper I'll be presenting: https://arxiv.org/abs/2302.02941 Sorry for the last-minute change, but I think it's for the best! See you later Best, Pablo

2 4

Wednesday reading group
by Adrián Javaloy 05 Mar '24

05 Mar '24

Dear all, This week's reading group is hosted by *udS*, and we will have *Pablo Sanchez* talking about the quite recent paper Statistics without Interpretation: A Sober Look at Explainable Machine Learning <https://arxiv.org/abs/2402.02870>. You can find an abstract of the paper below. As always, the reading group will be on *Wednesday at 2:30pm at room 0.01 of the E1.7 building*. I am in the process of find a free room with a working projector, so I will update you if the room changes at some point, for now we meet where always. If someone plans to join remotely, please *let me know in advance.* *Abstract* In the rapidly growing literature on explanation algorithms, it often remains unclear what precisely these algorithms are for and how they should be used. We argue that this is because explanation algorithms are often mathematically complex but don't admit a clear interpretation. Unfortunately, complex statistical methods that don't have a clear interpretation are bound to lead to errors in interpretation, a fact that has become increasingly apparent in the literature. In order to move forward, papers on explanation algorithms should make clear how precisely the output of the algorithms should be interpreted. They should also clarify what questions about the function can and cannot be answered given the explanations. Our argument is based on the distinction between statistics and their interpretation. It also relies on parallels between explainable machine learning and applied statistics. Cheers, Adrián

2 1

Cancelled this week's RG
by Adrián Javaloy 14 Feb '24

14 Feb '24

Dear all, Unfortunately, we need to cancel this week's reading group, as our group will be busy carrying out the interviews for new PhD candidates. If both groups are available (still to be confirmed) we will have this week's reading group next week instead. Otherwise, we will resume in two weeks as usual. Thanks for your understanding. I will keep you updated. Have a nice week, Adrián

1 2

[ML RG] Wednesday 2:30pm @ CISPA 0.07
by Siu Lun Chau 24 Jan '24

24 Jan '24

Dear all, For this week’s ML reading group, we have Dr. Siu Lun Chau presenting the paper titled Distribution-Free Uncertainty Quantification for Kernel Methods by Gradient Perturbations<https://arxiv.org/abs/1812.09632>. It’s an interesting work providing an alternative way of quantifying uncertainty with no distributional assumption on the noise term (unlike standard GPs). This time the meeting will take place on Wednesday 2:30pm at CISPA building room 0.07. There should be cookies and some drinks this time. Here’s the zoom link for those who cannot attend physically: https://cs-uni-saarland-de.zoom.us/j/91937148099?pwd=R1BhUVVVMXJKRWU1OXI5al… Best regards, Siu Lun Chau --- Siu Lun (Alan) Chau, Ph.D (Oxon) | Postdoctoral Researcher CISPA - Helmholtz Center for Information Security Stuhlsatzenhaus 5, 66123 Saarbrücken, +44 7415137484 | siu-lun.chau(a)cispa.de | https://chau999.github.io/

1 1

Reading Group Wednesday 10th
by Adrián Javaloy 10 Jan '24

10 Jan '24

Dear all, I hope you had a pleasing Christmas break, and that you had time to charge up your batteries and spend time with family and friends. This week we are back with the ML reading group and *Georgi Vitanov*, a new Master student at Isabel's group, will be introducing the paper titled Causal Fairness under Unobserved Confounding: A Neural Sensitivity Framework <https://arxiv.org/abs/2311.18460>. As usual, the reading group will take place on *Wednesday* at *2:30pm*, and this time UdS will be hosting the meeting at room *0.001* of the building *E1.7* (the cluster of excellence building, the room is the first one on the left as you enter from the front door of the building). If you are unable to attend physically, a Zoom call will be available in the same link as always: https://cs-uni-saarland-de.zoom.us/j/91937148099?pwd=R1BhUVVVMXJKRWU1OXI5al… Cheers, Adrián

1 2

Reading group on December 11th
by Batuhan Koyuncu 13 Dec '23

13 Dec '23

Dear all, For this week's reading group, we are most probably doing it online given that I have covid. I will email you again on Wednesday after testing on the same day. Best, Batu

1 1

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

2024

2023

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

Ml-reading-group