6

A question that has been bothering me for so long about research work is, what level of work can be called research and is worthy of publishing in a research paper?

Let's say I am working with JPEG compression, if I make some may be a little enhancement in the algorithm for the compression, can it be given the title of research.

So my question is:

If I make a little enhancement or added a little new stuff to an already existing work, can it be called a research and can I write a research paper for it?

0
3

The question I would ask is: "Are there people working in this area who would learn something interesting and useful from what you've done?" If the answer is yes then your paper makes a contribution and so is potentially publishable. At that point you could consider some secondary questions, e.g.:

  • Length of the paper: this should be commensurate with the extent of the contribution (i.e., how much someone would learn) -- a small contribution merits a short paper.

  • Venue of publication: some place (conference, journal, forum) typically accessed by the people who would learn the most from your paper. For example, a theory journal would not be the right place for systems-oriented paper about the guts of a compiler.

2

What matters most is the time and memory complexity of the algorithm. The crucial step is having the improvements repeatable (reproducible) by another person.

E.g. if instead of O(n2) your algorithm offers O(n log(n)) execution time, it is worth publishing and you may well expect your name to be remembered for a long time.

If your implementation of the algorithm has lower startup time, or spends 5% less time on each iteration, it is not altogether clear that this is an advance in the field. With another compiler or with error-checking on, these results may not be valid.

There are important exceptions: self-tunable, configurable, hardware error-resistant, hard realtime algorithms, to name a few. Adding valuable properties to algorithms is a very nice result.

1
  • I don't think this is necessarily true as the area of improvement depends upon what sort of area you are working in. For example in a lot of CS areas time and memory complexity are secondary issues and accuracy might be primary. Sep 4 '13 at 3:26
2

I would definitely say yes, if you researched your field thoroughly and found a solution to a problem, why not? But just to make sure: I would discuss my ideas with a supervisor, or colleague first. A lot of people made improvements on existing algorithms and wrote about it. I have read a few papers about compiling mpeg on the GPU using OpenCL.

Check out Erich Marths Paper: "Parallelization of the x264 encoder using OpenCL" at ACM he did improve some aspects, but interesting is rather what was researched. We tried his code and approach years in 2012 to find: CPUs had outgrown the improvement already and the approach could not be followed any more without significant work put into it.

Just make sure your research effort becomes clear. If in doubt, hand the draft to a "mentor" or colleague to judge the degree of research effort.

2
  • doesnt the ammount of enhancement matter ? !
    – Ali
    Sep 3 '13 at 9:17
  • 1
    not too much. Maybe you just found the tool xy doesn't serve the purpose you hoped and therefore found that it was interesting research, but cannot make any significant improvement.
    – AnyOneElse
    Sep 3 '13 at 10:01
1

Actually, you are not necessarily required to show any improvement in performance.

What journals typically require of a publication is for the results to be novel, sound and withing the scope of the journal's area. More prestigious journals also judge how interesting your results are to their readership. However, some journals will accept any work as long as it is novel and technically sound (e.g. PLoS ONE).

As an example, if you find an algorithm that has performance that is similar to current state-of-the-art algorithms but gives some new insight about the structure of the problem it could be interesting enough for publication.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.