As part of ILP 2016, we are holding the inaugural competition on Inductive Logic Programming. This is designed to test the accuracy, scalability and versatility of the learning systems which are entered. There are two main tracks for the competition: probabilistic and non-probabilistic. The winner in each of these tracks will be invited to submit an extended paper on the problem domain and their solution to the ILP'16 special issue of the Machine Learning journal.
Due to a lack of entrants, we decided to relaunch the compeition in November 2016.
All participants are required to register on the competition website. Once they have registered, they will be able to upload their system along with a script to interface with the competition (translating the competition input into an input for their system). All participants should also submit a short document (maximum 3 pages) explaining how their implementation works. This document will appear on the competition website after the competition (their implementation will remain private). Between November 2016 and February 2017, participants may upload new versions of their system as often as they like. Whenever a new version is uploaded, it will be run on a set of specimen datasets, which reflect the range of datasets used in the final competition. They will be given a full output from the server on how they would have scored in the competition on these specimen datasets.
This phase will allow them to ensure that their implementation handles the competition input with no problems and also allow them, if they wish, to try to tailor their algorithm/configuration for the problem domain in the competition.
There are two main tracks for the competition: probabilistic and non-probabilistic. A set of specimen problems from both tracks are available for download.
The final competition will be run on February 8th 2017. This will be with a larger number of (entirely new) data sets. Although the data sets will be new, the problem domain will remain the same. The results will be announced on February 22nd 2017. Immediately after results have been announced, all datasets will be published on the competition website.
The competition has now finished. Although there was only a single entry in both the probabilistic and non-probabilistic track, this should not detract from the performance of the two systems, or the effort made by the authors. We would like to congratulate the two winners. Details on the final competition problems and the performance of the two systems on these problems can be found by following the links in the table below.
|Non-probabilistic:||Peter Schüller||Marmara University||For more details on this system and its performance, see here.|
|Probabilistic:||Riccardo Zese, Elena Bellodi and Fabrizio Riguzzi||University of Ferrara||For more details on this system and its performance, see here.|
The datasets used in the competition are now available. They can be downloaded from the links in the table below. The initial datasets were available before the competition and were used by the competition entrants to tailor their systems, whereas the final datasets were unseen and were used in the final competition.
|Track||Initial dataset||Final dataset|
|Non-probabilistic:||Prolog / ASP||Prolog / ASP|
|Probabilistic:||Prolog / ASP||Prolog / ASP|
Any issues with the datasets above should be reported to Mark Law (firstname.lastname@example.org)