Full Draft

[dashaasienga]

@katcorr

I hope your weekend was great!

Following our plan (#25), I've pushed what is essentially a full draft of the thesis (sans acknowledgments) here: https://github.com/dashaasienga/Statistics-Senior-Honors-Thesis/blob/main/Thesis/index/_book/Dasha-Asienga_StatThesis.pdf!

I'd love your feedback on it,

1) mostly focusing on the newer sections: Chapter 4, conclusion, and abstract 2) and then, on the thesis body as a whole!

I'd like to send you back the revised Chapter 4, conclusion, and abstract (with your edits) by Thursday this week (since those are the newer sections that I'd like to have at least a second pass on) and then work on any adjustments to the organization and flow of the entire thesis body itself afterward.

The conclusion page is feeling a bit long right now (~3 pages), so I'd like to continue exploring ways to shorten that and the abstract is basically a 200-word summary of the conclusion.

See you on Tuesday!

[katcorr]

Hi! Great progress! I added comments to the draft, which you can find here. A few bigger picture thoughts:

• abstract: I didn't end up adding more comments to the pdf since we discussed in our meeting and you took notes on those suggestions about reformatting so it doesn't state "In Chapter . . . " :)
• Ch. 4 results: beautifully organized and clear presentation of results
• conclusion: I don't think the conclusion is too long. However, I think some parts of it could be shortened and others given more space
  • give (a little) less space: paragraphs 2-5 when you summarize each chapter. As with the abstract, I don't usually seem the Conclusion stating a summary of each Chapter in that way ("In Chapter 1, ..."); however, I do kind of like how you did that in that, there's a ton of information, and I like how it directs me to which chapter I should review if I wanted more info about that part. That being said, is there a way you could cut back on these paragraphs so it's pulling out only the key 1-2 primary ideas or "takeaways"?
  • give (a little) more space: the second to last paragraph could be elaborated on. Or maybe not so much elaboration, as a restructuring. It seems to combine a summary and also a list of future steps. I think you could separate out the future directions to each have their own sentence instead of one long sentence trying to name multiple different avenues for future work in one sentence.

In conclusion, although a step in the right direction toward fair ML, the Seldonian framework is far from the perfect solution. Seldonian algorithms may still yield unfair results, and even producing fairer outcomes, there is a practical limit to how much fairness can be enforced while retaining reasonable model predictive performance. In line with such concerns throughout the fair ML landscape, Microsoft’s Fairlearn is designed to help navigate trade-offs between fairness and model performance (Bird et al., 2020). Thus, continued work involves investigating similar ways to optimize fairness and predictive performance within the Seldonian framework and find a Paretooptimal solution, assessing Seldonian performance in practical continuous settings, exploring how subgroup or individual notions of fairness can fit into the Seldonian framework as well as their performance in comparison to the group notions that were the focus of this research, and finally, comparing Seldonian outcomes with other state-of-the-art fair ML tools such as Microsoft’s Fairlean and IBM’s Fairness 360.

[dashaasienga]

Thanks for your suggestions both here and in issue #37!

I made a few small edits in Chapter 4, and the only place where I added some major revisions was Section 4.3. I also made some significant revisions to both the conclusion and abstract. In line with the plan (#25), I've pushed the latest version: https://github.com/dashaasienga/Statistics-Senior-Honors-Thesis/blob/main/Thesis/index/_book/Dasha-Asienga_StatThesis.pdf.

I recall you mentioned you have some time tomorrow! For this revision, I think it'll be most important to focus on the thesis body as a whole, and especially the abstract and conclusion, since those are the individual parts where I did some re-structuring and are also the most important for capturing a snapshot of the entire thesis paper. I'll spend the break working on those edits and revisions and also just making sure everything looks and feels good-ish in time for the Wednesday due date! :)

Looking forward to your feedback!

[dashaasienga]

One bigger-picture thing I noticed when I read through the entire draft was in Chapter 2, I used $m$ to refer to sample size and $n$ to refer to the number of behavioral constraints. I never had to keep track of the number of behavioral constraints after that because we only used 1, so later chapters use $n$ for sample size as is more typical. I think the distinction between $m$ and $n$ is clear in Chapter 2 itself, but I'm wondering whether it may be confusing for continuity and integration with other chapters. If that's the case, I can switch to using $m$ for the number of behavioral constraints and $n$ for sample size consistently throughout. Let me know what you think of this!

[katcorr]

Oh my gosh, the thesis has come together SO well. It really reads beautifully. Here are my (all minor!) edits after reading through the full draft.

• The reorganized abstract reads much better. I only had minor edits to it.
• The introduction is strong as well. For the first section, I suggested adding some citations. I don't know that this is really necessary in this section, but conversations keep arising at the college about concerns around lack of citations so perhaps I'm erring on the side of caution here. I could check with other faculty to see what they think and get back to you if you want.
• Speaking of citations, double check your references format. Some seem like they need more info (e.g., if they're a website instead of a journal, add the URL and date last accessed)
• I saw your message about $m$ and $n$. Yes, I agree with changing the number of constraints to $m$ so you can consistently refer to sample size as $n$
• The conclusion also is solid! Only minor edits there.

GREAT WORK! YAY! Enjoy the break :)

[dashaasienga]

THANK YOU!!

I revised the draft, but I have 2 pending questions (which we can also look into tomorrow):

• [NEVER MIND!! I needed to switch to @Electronic, so all good] In terms of references, I added the URLs for the websites, but I'm having a lot of trouble adding the date last accessed field in the BibTex format. I've spent the past ~1 hour trying different field options and loading different bib-latex packages, but nothing seems to be working, so I was wondering if you have any idea of how to fix this from past theses. Online, it seems this is the most recommended way to add the date last accessed, but it just doesn't render when I knit:

• During my exploration of the Tex files, I discovered we can change the stylistic format of the thesis body, including the font. I was wondering if the format we have now is the standard, or if there is room to play around with different fonts, etc. Not that it makes a huge difference, but I was just curious :)