Before submitting your paper to Evolutionary Computation, please read both the publisher's guidelines for authors, and the more dedicated text below.
ECJ accepts papers that broadly fall into the three categories Applications/Experimental Results/Theory described below. Of course, many papers may fall into more than one category.
(1) Applications: a commonly accepted idea is that Evolutionary Computation only focuses on the fundamentals of bio-inspired algorithms. This is not at all the case, and application papers describing original work are welcome. A minimal requirement, though, is to carefully present the application domain, assuming that the reader is not familiar with even its basic concepts.
It is not sufficient simply to apply an arbitrarily chosen EA to a new problem domain and show some results. An application paper will likely fall into one of two categories:
· Results oriented: contain strong evidence that the use of Evolutionary Computation resulted in break-through results that could not be achieved by another method
· Methodological: papers where the specificities of the real-world problem led to significant algorithmic improvements that could possibly be generalized to other areas
Such "success stories" are of course reviewed by experts in the application domain, i.e. outside our usual pool of reviewers.
(2) Experimental results: as in all experimental sciences, the first requirement is that of reproducibility. All details of the algorithm, parameter values, etc. must be clearly given - and remember that there is no such thing as "a standard EA" that would replace the complete description of your algorithm. Moreover, because we are dealing with stochastic algorithms, strong conclusions drawn from experimental results must rely on statistically sound analysis. There have been to-date many tutorials, workshops and books describing statistically sound experimental protocols. Although all papers are different, please consider the following guidelines when preparing your manuscript:
· Ensure reproducibility (pseudo-code, parameters etc.)
· Use the most recent and well-cited benchmarks to evaluate your work. Show the generality applicability of your method across a broad a range of instances as possible – there are many datasets freely available – testing on a small selected instance set is usually not sufficient.
· Provide a comparison to state-of-the-art results from both EC methods and other fields if appropriate
· Provide evidence of tuning rather than arbitrary selection of parameters (there are many packages for doing this for you)
· Consider providing code as supplementary data
· Consider providing datasets if possible as supplementary data
· Provide a proper statistical analysis. This goes beyond a table showing mean/standard deviation. Use appropriate parametric/non-parametric statistics to show statistical significance with stated confidence levels
(3) Theoretical results: submitted material should include all proofs. While conference papers, because of space limitation, often cannot present the complete proofs, journal papers must be self-contained, or refer to easily accessible published results when citing existing proofs (e.g. no "Personal Communication"). In exceptional cases (e.g. Andrew Wiles' proof of Fermat theorem), parts of the proofs can be added as supplementary material.
Because reviewers are a scarce resource, submitted papers that fall short on some of the issues above will be rejected without entering the usual journal review process.
General Submission Guidelines for all papers:
Manuscript Preparation & Submission
Guidelines updated from Marc Schoenauer's Editorial Introduction of Evolutionary Computation 14(4) - Winter 2006