Another data competition:
Machine Learning for Signal Processing (MLSP) TC Announces the Winners of the 6th Annual Data Analysis Competition
See here for more info.
Archive for the 'reproducible research' Category
Making publications reproducible is tough…
I recently experienced it again in some of my work. In the stress of preparing a publication for a submission deadline, it is very challenging to take the (precious) time to verify all of the results once more and make sure all the results are perfectly reproducible. A result or figure so easily slips in for which the exact parameter settings have not been checked or written down…
I just got pointed to the author guidelines for CVPR 2010. They state that reviewers will be asked about (indicative) reproducibility (or repeatability, as it is called there):
Repeatability Criteria: The CVPR 2010 reviewer form will include the following additional criteria, with rating and associated comment field: “Are there sufficient algorithmic and experimental details and available datasets that a graduate student could replicate the experiments in the paper? Alternatively, will a reference implementation be provided?”. During paper registration, authors will be asked to answer the following two checkbox questions: “1. Are the datasets used in this paper already publicly available, or will they be made available for research use at the time of submission of the final camera-ready version of the paper (if accepted)? 2. Will a reference implementation adequate to replicate results in the paper be made publicly available (if accepted)?” If either these boxes are checked, the authors should specify in the submitted paper the scope of such datasets and/or implementations so that the reviewers can judge the merit of that aspect of the submission’s contribution. The Program Chairs realize that for certain CVPR subfields providing such datasets, implementations, or detailed specification is impractical, but in other areas it is reasonable and sometimes even standard, so on balance repeatability is a relevant criteria for reviewer consideration. “N.A.” will be an available reviewer score for this field, as it is for other fields.
Very exciting developments!
Keith Baggerly and Kevin Coombes have just published a great paper about forensic bioinformatics: “Deriving chemosensitivity from cell lines: forensic bioinformatics and reproducible research in high-throughput biology.” The article will appear in the next issue of Annals of Applied Statistics and is available here.
The blog post Make up your own rules of probability discusses a couple of the innovative rules of probability Baggerly and Coombes discovered while reverse engineering bioinformatics articles.
Note that the Reproducible Ideas blog is winding down. I’m in the process of handing over the reproducibleresearch.org domain to the owners of reproducibleresearch.net. Eventually the .net site will move over to the .org name.
Some more interesting reading:
K. Price, Anything You Can Do, I Can Do Better (No You Can’t)…, Computer Vision, Graphics, and Image Processing, Vol. 36, pp. 387-391, 1986, 
doi:10.1016/0734-189X(86)90083-6.
Abstract: Computer vision suffers from an overload of written information but a dearth of good evaluations and comparisons. This paper discusses why some of the problems arise and offers some guidelines we should all follow.
Very nice reading material, and (although I know these ideas are around for quite some time already) I was amazed to see so many parallels to our recent IEEE Signal Processing Magazine paper, already in this paper by Price from 1986. That’s more than 20 years ago! Price talks about the reproducibility problems in computer vision and image processing, writing we should “stand on other’s shoulders, not on other’s toes”. He also did a study on reproducibility of a set of about 42 papers, verifying the size of the dataset and clarity of the problem statement. Price concludes as follows: “Researchers should make the effort to obtain implementations of other researchers’ systems so that we can better understand the limitations of our own work.”
Again, interesting to see how these issues and worries have been around for more than 20 years in the field of image processing. It’s about time to drastically improve our standards, I think!
I would really recommend this article to anyone interested in issues around reproducible research.
I just read the following paper:
A. J. Rossini and F. Leisch, Literate statistical practice, UW Biostatistics Working Paper Series 194, University of Washington, WA, USA, 2003.
Although I am not a statistician, this was a very interesting paper to me. It gives a nice description of a possible literate programming approach in statistics. The authors propose a very versatile type of document combining documentation and code/statistical analyses, interweaved as in the original description of literate programming by Knuth. From this versatile document, which contains a complete description of the research work, multiple reports can be extracted, such as an article, an internal report, an overview of the various analyses that were performed, etc.
I just read the article about Netflix’ Million Dollar Programming Prize on IEEE Spectrum.
Robert M. Bell, Jim Bennett, Yehuda Koren, and Chris Volinsky, The Million Dollar Programming Prize, IEEE Spectrum Online, http://www.spectrum.ieee.org/may09/8788.
Interesting article, showing again how contests proposing a challenge can inspire a lot of great work, and allow an ‘objective’ comparison between algorithms. I think they provide a great way to motivate researchers to work on real problems, with testing on standardized datasets.
I am glad to let you know that our paper has been published in the latest issue of IEEE Signal Processing Magazine:
P. Vandewalle, J. Kovacevic and M. Vetterli, Reproducible Research in Signal Processing – What, why, and how, IEEE Signal Processing Magazine, Vol. 26, Nr. 3, pp. 37-47, 2009, DOI: 10.1109/MSP.2009.932122.
Have you ever tried to reproduce the results presented in a research paper? For many of our current publications, this would unfortunately be a challenging task. For a computational algorithm, details such as the exact data set, initialization or termination procedures, and precise parameter values are often omitted in the publication for various reasons, such as a lack of space, a lack of self-discipline, or an apparent lack of interest to the readers, to name a few. This makes it difficult, if not impossible, for someone else to obtain the same results. In our experience, it is often even worse as even we are not always able to reproduce our own experiments, making it difficult to answer questions from colleagues about details. Following are some examples of e-mails we have received: “I just read your paper X. It is very completely described, however I am confused by Y. Could you provide the implementation code to me for reference if possible?” “Hi! I am also working on a project related to X. I have implemented your algorithm but cannot get the same results as described in your paper. Which values should I use for parameters Y and Z?”
Enjoy reading! And feel free to post your comments!
Last month, a few former colleagues at LCAV did some cross-testing of the reproducible research compendia available at rr.epfl.ch. And I must say, from the results I have seen so far, it is quite a sobering experience. Many of those which I considered to be definitely reproducible didn’t pass the test (entirely). I guess that shows again how difficult it is to make work really reproducible, even if you fully intend to do it. So that also leads me to my conviction that for papers that do not have code and data online, it is almost impossible to reproduce the exact results. There is work to be done on the road to reproducible research!
I’ll need to look further into the reasons why even some of my own work did not pass the test.
On doing research
I was just reading the following two articles/notes. While they are not entirely about reproducible research, I think they reflect well the worries that many researchers have about current “publish or perish” research practices. Not sure I agree with all of it, but they do make a number of good remarks.
D. Geman, Ten Reasons Why Conference Papers Should be Abolished, Johns Hopkins University, Nov. 2007.
Y. Ma, Warning Signs of Bogus Progress in Research in an Age of Rich Computation and Information, ECE, University of Illinois, Nov. 2007.