Artifact Evaluation for Computer Systems' Research
We work with the community and ACM to improve methodology and tools for reproducible experimentation, artifact submission / reviewing and open challenges!
Home Artifacts Joint Committee Submission Guide Reviewing Guide FAQ Prior AE

Updated guide is now available here



This guide (V20160509) was prepared by Grigori Fursin and Bruce Childers with contributions from Michael Heroux, Michela Taufer and other colleagues to help you describe and submit your artifacts for evaluation across a range of CS conferences and journals. It gradually evolves based on our long-term vision (TRUST'14@PLDI'14 and DATE'16) and your feedback after our past Artifact Evaluations (see AE CGO-PPoPP'17 discussions). It should also help you prepare your artifacts for a possible public release, if you plan to do so (for example as an auxiliary material in a Digital Library or on your personal web page).

Navigation:

What to expect

We are trying to make artifact submission as simple as possible. You just need to pack your artifact (code and data) using any publicly avialable tool you prefer. In some exceptional cases when rare hardware or proprietary software is used, you can arrange a remote access to machine with the pre-installed software.

Then, you need to prepare a small and informal Artifact Evaluation appendix using our AE LaTeX template (common for PPoPP, CGO, PACT and SC conferences - see below) to explain evaluators what are your artifacts and how to use them (you will be allowed to add up to 2 pages of this Appendix to your final camera-ready paper). Please see this PPoPP'16 paper for the example of such AE appendix.

At least two reviewers will follow your guide to replicate your results (for example, exact output match) or reproduce them (for example, varying performance numbers or scalability on a different machine), and will then send you a report with the following overall assessment of your artifact based on our reviewing guidelines:

where "met expectations" score or above means that your artifact successfully passed evaluation and will receive a stamp of approval (added to the paper itself):
The highest ranked artifact (usually not only reproducible but also customizable and reusable) will also receive a "distinguished artifact" award. This section is also used as a discussion forum with the community about how to improve AE.

Since our eventual goal is to promote artifact validation and sharing (rather than naming and shaming problematic artifacts), you will be able to address raised issues during the rebuttal. Furthermore, we allow a small amount of communication between reviewers and authors whenever there are installation/usage problems. In such cases, AE chairs will serve as a proxy to avoid revealing reviewers' identity (the review is blind, i.e. your identity is known to reviewers since your paper is already accepted, but not vice versa).

Preparing artifacts for submission

You just need to perform the following 4 steps to submit your artifact:
  1. Pack your artifact (code and data) or provide an easy access to them using any publicly available and free tool you prefer or strictly require.

    For example, you can use the following:
    • Virtual Box to pack all code and data including OS (typical images are around 2..3GB. we strongly recommend to avoid images larger than 10GB).
    • Docker to pack only touched code and data during experiment.
    • Standard zip or tar with all related code and data, particularly when artifact should be rebuilt on a reviewers machine (for example to have a non-virtualized access to a specific hardware).
    • Private or public GIT or SVN.
    • Arrange a remote access to machine with pre-installed software (exceptional cases when rare hardware or proprietary software is used or the VM image is too large)) - you will need to privately send the access information to the AE chairs. Also, please avoid making any changes to the remote machine during evaluation (unless explicitly agreed with AE chairs) - you can do it during rebuttal phase, if needed!
    You can also check and use other public tools available for artifact sharing.

    From our past Artifact Evaluation experience, we've noticed that the most challenging part is to automate and customize experimental workflows. It is even worse, if you need to validate experiments using latest software environment and hardware (rather than quickly outdated VM and Docker images). Most of the time, some ad-hoc scripts are used to implement these workflows. They are very difficult to change and customize, particularly when an evaluator would like to try other compilers, libraries and data sets.

    These problems motivated us to develop Collective Knowledge Framework (CK) - a small and open-source infrastructure and repository to help researchers share their artifacts as reusable Python components with a unified JSON API. This approach should help researchers quickly prototype experimental workflows (such as multi-objective autotuning) from such components while automatically detecting and resolving all required software or hardware dependencies. CK is also intended to reduce evaluators' burden by unifying statistical analysis and predictive analytics (via scikit-learn, R, DNN), and enabling interactive reports. Please see examples of a CK-based live repository, interactive article and PLDI'15 CLSmith artifact shared in CK format. Feel free to use it when submitting artifacts for the evaluation.

  2. Write a brief artifact abstract to informally describe your artifact including minimal hardware and software requirements, how it supports your paper, how it can be validated and what is the expected result. It will be used to select appropriate reviewers.

  3. Fill in and append AE template (download here) to the PDF of your accepted paper. Though it should be relatively intuitive, you can check out extra notes about this template based on our past AE experience.

  4. Submit artifact abstract and new PDF at the AE EasyChair website.
If you encounter problems, find some ambiguities or have any questions, do not hesitate to contact AE chairs of your conference or AE steering committee!

If accepted

You can now add the following stamp to the final camera-ready version of your paper: While there are no strict formatting rules for the stamp, please add it anywhere close to the title. For example, see PPoPP'15 article together with this LaTeX example. You can change \hspace and \raisebox parameters to better fit stamp to your paper.

We strongly encourage you to submit your AE appendix (up to 2 pages) as an auxiliary material for the Digital Library (while removing all unnecessary or confidential information) along with the final variant of your paper. This will help readers better understand what was evaluated.

Though you are not obliged to publicly release your artifacts (in fact, it is sometimes impossible due to various limitations), we also strongly encourage you to share them with the community (even if they are not open-source). You can release them as an auxiliary material in Digital Libraries together with your AE appendix or use your institutional repository and various public services for code and data sharing.

Even accepted artifacts may have some unforeseen behavior and limitations discovered during evaluation. Now you have a chance to add related notes to your paper as a future work (if you wish)..

Note, that we are developing open-source Collective Knowledge framework to help researchers share their artifacts as unified, reusable and customizable components together with experimental workflows and interactive articles. If you are interested to know more, please check some online examples and a recent DATE'16 paper.

A few examples of accepted artifacts from the past conferences, workshops and journals

Methodology archive

We keep track of all past versions of submission/reviewing methodology to let readers understand which one was used in papers with evaluated artifacts.

Thank you for participating in Artifact Evaluation!


This guide was prepared by Grigori Fursin and Bruce Childers with contributions from Michael Heroux, Michela Taufer and other colleagues.
Maintained by
cTuning foundation (non-profit R&D organization)
and volunteers!
          
Powered by Collective Knowledge
                     
  
  
  
           Locations of visitors to this page