Writing Technical Articles

Read Strunk and White, Elements of Style. Again.

Give the paper to somebody else to read. If you can, find two people: one person familiar with the technical matter, another only generally familiar with the area.

Papers can be divided roughly into two categories, namely original research papers and survey papers. There are papers that combine the two elements, but most publication venues either only accept one or the other type or require the author to identify whether the paper should be evaluated as a research contribution or a survey paper. (Most research papers contain a "related work" section that can be considered a survey, but it is usually brief compared to the rest of the paper.)

Research Papers

A good research paper has a clear statement of the problem the paper is addressing, the proposed solution(s), and results achieved. It describes clearly what has been done before on the problem, and what is new.

The goal of a paper is to describe novel technical results. There are four types of technical results:

  1. An algorithm;
  2. A system construct: such as hardware design, software system, protocol, etc.;
    One goal of the paper is to ensure that the next person who designs a system like yours doesn't make the same mistakes and takes advantage of some of your best solutions. So make sure that the hard problems (and their solutions) are discussed and the non-obvious mistakes (and how to avoid them) are discussed. (Craig Partridge)
  3. A performance evaluation: obtained through analyses, simulation or measurements;
  4. A theory: consisting of a collection of theorems.
A paper should focus on

Paper Structure

It is recommended that you write the approach and results sections first, which go together. Then problem section, if it is separate from the introduction. Then the conclusions, then the intro. Write the intro last since it glosses the conclusions in one of the last paragraphs. Finally, write the abstract. Last, give your paper a title.



Body of Paper



Things to Avoid

Guidelines for Experimental Papers

"Guidelines for Experimental Papers" set forth for researchers submitting articles to the journal, Machine Learning.
  1. Papers that introduce a new learning "setting" or type of application should justify the relevance and importance of this setting, for example, based on its utility in applications, its appropriateness as a model of human or animal learning, or its importance in addressing fundamental questions in machine learning.
  2. Papers describing a new algorithm should be clear, precise, and written in a way that allows the reader to compare the algorithm to other algorithms. For example, most learning algorithms can be viewed as optimizing (at least approximately) some measure of performance. A good way to describe a new algorithm is to make this performance measure explicit. Another useful way of describing an algorithm is to define the space of hypotheses that it searches when optimizing the performance measure.
  3. Papers introducing a new algorithm should conduct experiments comparing it to state-of-the-art algorithms for the same or similar problems. Where possible, performance should also be compared against an absolute standard of ideal performance. Performance should also be compared against a naive standard (e.g., random guessing, guessing the most common class, etc.) as well. Unusual performance criteria should be carefully defined and justified.
  4. All experiments must include measures of uncertainty of the conclusions. These typically take the form of confidence intervals, statistical tests, or estimates of standard error. Proper experimental methodology should be employed. For example, if "test sets" are used to measure generalization performance, no information from the test set should be available to the learning process.
  5. Descriptions of the software and data sufficient to replicate the experiments must be included in the paper. Once the paper has appeared in Machine Learning, authors are strongly urged to make the data used in experiments available to other scientists wishing to replicate the experiments. An excellent way to achieve this is to deposit the data sets at the Irvine Repository of Machine Learning Databases. Another good option is to add your data sets to the DELVE benchmark collection at the University of Toronto. For proprietary data sets, authors are encouraged to develop synthetic data sets having the same statistical properties. These synthetic data sets can then be made freely available.
  6. Conclusions drawn from a series of experimental runs should be clearly stated. Graphical display of experimental data can be very effective. Supporting tables of exact numerical results from experiments should be provided in an appendix.
  7. Limitations of the algorithm should be described in detail. Interesting cases where an algorithm fails are important in clarifying the range of applicability of an algorithm.

The Conference Review Process

It is hard to generalize the review process for conferences, but most reputable conferences operate according to these basic rules:

  1. The paper is submitted to the technical program chair(s).
  2. The TPchair assigns the paper to one or more technical program committee members, hopefully experts in their field. The identity of this TPC member is kept secret.
  3. The TPC member usually provides a review, but may also be asked to find between one and three reviewers who are not members of the TPC. They may be colleagues of the reviewer at the same institution, his or her graduate students or somebody listed in the references. The graduate student reviews can be quite helpful, since these reviewers often provide more detailed criticism rather than blanket dismissal. Any good conference will strive to provide at least three reviews, however, since conferences operate under tight deadlines and not all reviewers deliver as promised, it is not uncommon that you receive only two reviews.
  4. The TPChair then collects the reviews and sorts the papers according to their average review scores.
  5. The TPC, or usuaally a subset that can make the meeting, then meets Usually, the bottom third and the top third are rejected and accepted without (much) further discussion. The papers discussed are those in the middle of the range, where a TPC member feels strongly that the paper ended up in the wrong bin, or where the review scores differ significantly, in particular if there are only two reviews.

Other References


This page contains material provided by Gail Kaiser, Craig Partridge, Eric Siegel, Sal Stolfo, Luca Trevisan, Yechiam Yemini, Erez Zadok.
Last updated by Henning Schulzrinne