2/9/24: Discuss how to find a research project, etc.

 

---

 The Human Race Into Space Requires Kidneys 

  

          A research and discussion group             

---

Below: Video of the world's largest indoor ocean (in Biosphere II, Summer 2022, Tucson, AZ)

 

Agenda and Minutes

1. Updates/announcements/status reports

• Updates?

2. Reading and discussion

• Current space reading:
• We finished a reading! https://www.newthingsunderthesun.com/pub/4xnyepnn/release/9.
  

• This updates our discussion of general guidelines for doing good PhD research projects from last week. Maybe we should read through it and discuss?
  
• Three characteristics of a reasonable research project:
  
1. Novelty (This is typically established through a literature search/review.)
   
2. Significance
3. Doability
• The US National Science Foundation does not define what science is, but nevertheless does have influence on research. They have a related take on novelty and significance. In proposals they typically require statements of Intellectual Merit and Broader Impacts. These are related but not identical to novelty and significance. Other funding agencies do things differently.
  
• How to find a research topic:
• You should be interested in your topic
• Good idea to build on an area in which you have some background and expertise, if any 
• Find an article. Consider replicating and extending it. This sounds simple in principle, but in practice it has some risks. Just ask VK. Still, it does define a bite-sized first step one can take.
  
• Do a shortish project resulting in a conference paper. Then you may gain enough experience and knowledge to help decide on what to do next. (Build on it? Change direction? Etc.) Suggested by IE. This is another bite-sized first step, though a bigger bite.
  
• It can be a research project.
• It can be a (publishable) systematic review of the literature.
• Ask your advisor to give you a research topic, or perhaps they will assign you one even if you don't ask.
• Here are two possibilities: 
1. find a weighted average of Wright's law and Moore's law that best fits some data. It's not as easy as it might first appear, but surely must be doable.
2. Extend the work of Kodali & Berleant, Visual Question Answering (VQA) on Images with Superimposed Text, https://arxiv.org/abs/2307.02489.
   
• Research methodology courses (you could take one, or you could ask to sit in on one informally)
  
• IFSC 7399: ST: Research Methods in Inf Sci 
• SYEN 7101: Research Methodology
  
• INFQ 7322: Information Quality Theory. Students write parts of a future dissertation (instructor: E. Alikhachkina).
  
• Taking courses is not the main point of a PhD! See the Wikipedia article.
• Here are some generic questions to ask when reviewing articles (and videos):
• What is the source?
• What is the most significant advance in the human knowledge presented in the paper?
• Why is that advance important?
• What important questions arise from the paper for future research?
• What important questions would it be nice if the paper answered, but does not answer?
• What does the paper present that is novel (no one else has provided that before)?
• What is the relevance of the paper to our research goals?
• Anything else to add to this list?
  
• Writing hints
• Often, each passage begins with a reference to what the reader already knows or was presented earlier, and ends with the most important new content that the reader should learn from the passage. Everything else goes in the middle. This applies to sentences, paragraphs, sections, ..., books, ....
  
• ChatGPT, Claude.ai, and other generative AIs can turn a rough draft into a more polished one, or even turn rough notes and lists of points into a draft. But, and it's a big but, the result needs a careful read-and-edit to catch cases of repetitiveness, passages that read well but upon reflection don't actually mean much, and passages that read well at first but are actually incorrect. Generative AIs can make good assistants, but the user has to be careful to stay in the driver's seat!
• I asked claude.ai how to improve on the guidelines and hints above. Here is its response:
1. Here are some suggestions for improving the PhD research guidelines:
2. Add more specifics on assessing novelty, significance, and feasibility of potential projects. Provide sample questions to ask or metrics to evaluate.
3. Expand the research topic section - suggest looking at recent conferences/journals in one's field, identifying gaps or open problems for exploration.
4. On building off existing work - caution against strict replication without extension. Discuss ethical considerations.
5. For bite-sized projects - recommend presenting at research meetups/posters rather than just conferences to start.
6. On methodology courses - also suggest auditing advanced seminars in one's subfield. Highlight value of understanding research designs.
7. Emphasize that reviewing literature and writing reviews can further skills and ideas.
8. On generic article questions - add prompts about limitations, assumptions, considering alternative explanations.
9. Writing section could include advice on creating an outline, getting feedback on drafts, targeting writing to audience.
10. Note using AI writing assistants takes practice - set expectations on need for editing, rewriting.
11. Throughout - relate tips to student's specific field/department, make actionable with directives like "ask your advisor..." or "look for ideas in recent issues of...

Let me know if you would like me to elaborate on any specific suggestions!

•  More to be added as needed...
  

• Current readings on kidneys or space (or both), depending on interest and attendance.
• Current kidney reading: 
• The Ensembles of Machine Learning Methods for Survival Predicting after Kidney Transplantation, Tolstyak et al., 2021, https://www.mdpi.com/2076-3417/11/21/10380. In a previous meeting we got up to "A classifier based on a decision tree is a tree whose inner vertices are denoted by terms," and can start with that next time we do this reading.

• Future space readings. See the page of possible future readings!
  
• Attendees could suggest possible readings/viewings between now and next time, or we could take time now for each person to search for things and report back.

• Future kidney articles, some evaluated for future reading and some yet to be evaluated.
  
• https://towardsdatascience.com/boruta-explained-the-way-i-wish-someone-explained-it-to-me-4489d70e154a (not evaluated yet).
  
• Deep learning-based classification of kidney transplant pathology: a retrospective, multicentre, proof-of-concept study, Jesper Kers*, Roman D Bülow*, Barbara M Klinkhammer, Gerben E Breimer, Francesco Fontana, Adeyemi Adefidipe Abiola, Rianne Hofstraat, Garry L Corthals, Hessel Peters-Sengers, Sonja Djudjaj, Saskia von Stillfried, David L Hölscher, Tobias T Pieters, Arjan D van Zuilen, Frederike J Bemelman, Azam S Nurmohamed, Maarten Naesens, Joris J T H Roelofs, Sandrine Florquin, Jürgen Floege, Tri Q Nguyen, Jakob N Kather†, Peter Boor†, https://pubmed.ncbi.nlm.nih.gov/34794930/. 9/1/23 vote was 3 5/6.
• Artificial intelligence and kidney transplantation, Nurhan Seyahi, Seyda Gul Ozcan, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8290997. 9/15/23 vote was 4.0. 
• TC wrote: Authors Zhang et al. (2020) conducted a similar study for clear cell renal cell carcinoma survival prediction, however they worked with publicly available genetic data from the Genomic Data Commons Data Portal (National Cancer Institute, n.d.) and the Cancer Genome Atlas (National Center for Biotechnology Information, n.d.). Using this data as well as other clinical prognostic parameters, the authors identified five prognostic genes that they believe are better survival predictors of clear cell renal cell carcinoma than then TNM system. Considering that the authors only explored public data, there are potentially additional genes or non-publicly available variables that could be explored. [Reference:] Zhang, Z., Lin, E., Zhuang, H., Xie, L., Feng, X., Liu, J., & Yu, Y. (2020). Construction of a novel gene-based model for prognosis prediction of clear cell renal cell carcinoma. Cancer Cell International, 20(1), 27. https://doi.org/10.1186/s12935-020-1113-6
• Long-Term Time Series Forecasting and Updates on Survival Analysis of Glioblastoma Multiforme: A 1975–2018 Population-Based Study, https://doi.org/10.1159/000522611
• These describe a method for using recent lifetime data, which is the same problem that TC is working on: (1) Brenner, H., Gefeller, O., & Hakulinen, T. (2004). Period analysis for ‘up-to-date’ cancer survival data. European Journal of Cancer, 40(3), 326–335. https://doi.org/10.1016/j.ejca.2003.10.013. (2) Brenner, H., & Hakulinen, T. (2006). Up-to-date and Precise Estimates of Cancer Patient Survival: Model-based Period Analysis. American Journal of Epidemiology, 164(7), 689–696. https://doi.org/10.1093/aje/kwj243
• TC wrote: "On the other hand, authors Solé et al. (2014) investigate the dynamics of technological innovation and the conditions under which it may lead to explosive growth or enter a linear regime. It develops a generalized model of technological evolution, focusing on two crucial properties: the number of previous technologies required to create new innovations and the rate at which technology ages. The study explores two different models of combinatorial growth, one involving long-range memory and the availability of old inventions for new innovations, and the other with aging having a characteristic time scale" 
• Ref.: Solé, R. V., Amor, D. R., & Valverde, S. (2014). On singularities in combination-driven models of technological innovation (arXiv:1407.6890). arXiv. https://doi.org/10.48550/arXiv.1407.6890