Step 1

The most important things come first. We discuss the main message, delivered by a scientific paper. We explore the first three elements of a scientific paper: the abstract, the highlights, and the keywords. The main message shall reveal itself through all elements of the paper. But we leave the rest of it for the next time. Namely, the title, introduction, problem statement, goal of the computational experiment, and conclusion are left behind. We select a paper and exercise in the reconstruction of these three elements.

The seminar

1. The warm-up 3-minute test
2. Model, Algorithm, Method: Machine learning in a nut-shell
3. Step 0 homework results discussion
4. Step 1 homework, how to read: the scheme
5. Structure of the main message
6. Structure of the abstract
7. The second and the last slide of your talk
8. Extracting keywords
9. Highlights: compressing the paper
10. Instastructure for your homework
    • GitHub: organize the repository
    • LaTeX: compile your file and commit without temporary files
11. The papers to select from
12. Optional GPT-role discussion

Resources

Step 1 Youtube video (expected with online version)

Homework

1. Set up your GitHub repository using this template, see how
2. Select a paper to read from the list below
3. Reconstruct its
   1. Abstract
   2. Keywords
   3. Highlights
   4. Short motivation for why you selected this paper (no templates here, since it is an extra topic to discuss)
4. Compile and upload TEX and PDF to GitHub (no temporary files, please)
5. Linear models for the next warm-up test, either
   1. look for the terms dot product, scalar projection, least squares, linear map
   2. or do fun-reading, the pages 33-39 from the book Section L3.

Note that we always respect your credit hours. So please keep track of it.

Your profit here is your ability to find the main message of a paper.

Abstract

The abstract of a paper is the first piece the reader looks at. Usually, it is written at the beginning of research and after the paper is done, before submission. Due to its importance, several versions of the abstract from different points of view are welcome.

The abstract is limited to 600 characters. It may contain

1. wide-range field of the investigated problem,
2. narrow problem to focus on,
3. features and conditions of the problem,
4. the idea of the suggested solution,
5. the novelty and alternative solutions to compare with,
6. application to illustrate with.

Examples of abstracts to discuss, a draft.

Keywords

The keywords of your paper shall match the subject of your research, and show the area and the focus. Ensure these keywords are used in your paper frequently and play an important role. They shall be recognized terms in your field of knowledge. See detailed explanations and Elsevier recommendations.

Highlights

To write highlights, see elsevier official version, a useful piece of advice, and Medium clarifications.

Papers to choose from

Please read a paper from this list and formulate its main message. Imagine you are a journal editor or a reliever, who receives scientific papers randomly and pick up some paper.

IMPORTANT. Since the homework is to reconstruct the abstract of one of these papers, please, try to skip the published abstract. Cover it and start reading according to the discussed reading scheme.

1. Distinguishing time-delayed causal interactions using convergent cross mapping DOI
2. Comprehensive study of feature selection methods to solve multicollinearity problem according to evaluation criteria DOI, PDF
3. Spatio-temporal filling of missing points in geophysical data sets DOI
4. Analytic and stochastic methods of structure parameter estimation DOI
5. Longitudinal predictive modeling of tau progression along the structural connectome DOI
6. Generative or Discriminative? Getting the Best of Both Worlds PDF
7. Neural Ordinary Differential Equations NIPS, Appendix
8. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations DOI, GitHub
9. How much does it help to know what she knows you know? An agent-based simulation study DOI

Can I select a paper of my own choice?

Yes. Here are some formal requirements.

1. A clear message in the area of Theoretical Foundations of Machine Learning.
2. No Kaggle-style papers with messages like "It works, but nobody knows how".
3. Top peer-reviewed journals, no ArXiv, better avoid conferences.
4. No papers from other fields: linguistics, medicine, finance, physics, etc. There must be only one primary subject: Machine Learning.
5. No overviews of paper collections, it is another genre.
6. No predatory publishing houses

Recommended journals

1. Machine Learning
2. Expert Systems with Applications
3. Journal of Machine Learning Research
4. Artificial Intelligence
5. Neurocomputing
6. IEEE Transactions on Pattern Analysis and Machine Intelligence
7. Neural Networks
8. Pattern Recognition
9. Data Mining and Knowledge Discovery
10. Nature Machine Inlelligence, the problem is the first word here is Nature so it focuses on natural sciences

See also

1. Elsevier's open archive
2. Springer's open archive

Transcript of the video

Appears after the seminar.