Dear students,

You can now find your final grades for the Causethical seminar in CMS. They are displayed as "Points", but refer to your grade in the German grading scale. That is, 1.0 is the best grade and passing grade is 4.0. In case you have any questions regarding your grade please send us an email before 28th Feb (next Monday).

Cheers,

Causethical team

 

 

 

 

Dear all,

Tomorrow, we will have the last presentation of the robustness block followed by the panel session. For your reference, we will be discussing some of the following questions/points.

 

Questions about concepts related to robustness

• What is the difference between generalization and OOD generalization? What type of generalization do the models studied aim to achieve
• What is the difference between the tasks of domain adaptation and unsupervised domain adaptation? Which paper refers to which task?
• Is the meaning of "Invariant features" and "causal features" the same? What are the similarities and differences? Follow-up question: are all invariant features causal or vice versa?

 

Questions about the studied papers

• Why does IRM method fail? Is it due to the assumptions? is it the optimization? Are we modeling direct causes of actions or just invariant features?
• What type of prior knowledge is required in the IRM approaches? Do we need any causal knowledge at all, e.g. causal graph?
• How does the method proposed in the paper "Domain Adaptation by using..." relates/differs from IRM?
• Although deep CAMA trained on clean MNIST data generalizes better to unseen shifts and shows even improved performance when fine-tuned with shifted test data, it is the most complex model among the ones considered here. Therefore, is questionable whether it can be well-fitted in tasks more complex than MNIST. What are other limitations of CAMA?
• Regarding the last paper: In which way and for what could we use counterfactual statements to improve the approach?
• Regarding the last paper: What are the differences between (construction of) adversarial and real-world causal graphs?
• All the papers have a lot margin for improvement...which one would you choose to extend and why?

 

General robustness questions

• Are causally motivated robust machine learning approaches just a means to estimate the functional information on top of (given) causal structure? Or is there more to them?
• Should we focus on solving robustness problems through formulating vulnerability of models or through modeling perturbations and possible attacks? What are the advantages and drawbacks of each type of approach?
• Given the many non-trivial assumptions made in the various techniques, how can we measure progress in a way that is universally valid for the various techniques?
• Is not lack of robustness related to lack of fairness in algorithms? Both stem from spurious correlations learned. How are they theoretically different? will robust models be explainable and fair?
• Should we aim to develop robust machine learning in any application? Why?

 

General machine learning questions

• Looking at the results, the increase in performance does not seem to be as significant as the author claims. Is such a bold claim normal in research? (If you would like to learn about good academic research practice, I suggest reading https://arxiv.org/pdf/2108.02497.pdf)

 

 

See you tomorrow!

Best wishes,
Your CausethicalML Team

Dear students,

This is just a friendly reminder about the abstract submission for the block of robustness, which is today.  As with previous submissions, this is a short checklist with the information it should include: 

• The abstract should provide an overview of all the papers of the block with main conclusions comparing them and their views
• The abstract should be 1 page long using the template we provide in the materials section. This is 1 page excluding references (i.e., references can be on the next page).
• You should include references, whenever suitable/necessary.
• The abstract should include at least 2 questions that would be suitable to ask at the panel.

 

On a different note, please upload your presentation on CMS. Then, then send us an email (A) confirming that you have uploaded it on CMS and (B) whether you are ok that we upload your presentation under the materials section (visible to all students formally registered in this seminar), such that other's can profit from it. While we encourage you to share it with your peers, your decision will not influence your grading.

If you have any more questions, we are happy to answer them.

Looking forward to the last panel session!

Your Causethical ML Team

Hi all,

Thank you so much for your participation today and the engaging discussion! I personally very much enjoyed that so many of you entered the discussion with their ideas, examples and thoughts.

For those of you interested, I wanted to not leave you without a reference to two of the papers I mentioned today during the session:

• What’s Sex Got to Do With Fair Machine Learning? by Lily Hu & Issa Kohler-Hausmann (FAccT'21)
• Causal multi-level fairness by Vishwali Mhasawade & Rumi Chunara (AIES'21)

Enjoy the break! Keep safe and healthy! Looking forward to seeing you all again next year!

Best,

Miriam