Scalable Non-observational Predicate Learning in ASP
Authors: Mark Law, Alessandra Russo, Krysia Broda, Elisa Bertino
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This section presents an evaluation of Fast Non OPL5 on non OPL tasks, demonstrating that it outperforms other ASP-based ILP systems. In particular, it is significantly faster than ILASP [Law et al., 2015a]. The ILASP system consists of a collection of algorithms, each with various parameters. For both scenarios in this evaluation, we performed initial experiments to determine the best performing version of ILASP for the scenario and report results only for this version. |
| Researcher Affiliation | Academia | Mark Law1 , Alessandra Russo1 , Krysia Broda1 and Elisa Bertino2 1Imperial College London, UK 2Purdue University, USA |
| Pseudocode | Yes | Algorithm 1 abduce possibilities(T, e) pp = { , , {a. | a A} |A AS(bottom(B ectx))}; poss = ; while pp = do pp = ; for p pp do p exceptions = m except(T, e, p); for p exceptions do for + fix m fix+(T, e, p, ) do pp = pp { pinc pi + fix, , pas }; for fix m fix (T, p exceptions) do poss = poss { pinc pi , fix , pas }; pp = pp ; return poss; |
| Open Source Code | Yes | Fast Non OPL can be run by downloading the latest version of Fast LAS (currently 2.0.0) from https://spike-imperial.github.io/ Fast LAS/ and running Fast LAS with the --nopl flag. |
| Open Datasets | Yes | ILASP and Fast Non OPL were run on a set of 10 learning tasks, each consisting of 50 valid episodes and 50 invalid episodes (all learning tasks are available at https://github. com/spike-imperial/Fast LAS/blob/master/Fast LAS2/data). |
| Dataset Splits | Yes | The results9 of performing 10-fold cross validation are shown in Figure 2. |
| Hardware Specification | Yes | All experiments for Fast Non OPL, Fast LAS and ILASP were run on an Ubuntu 18.04 desktop machine with a 3.4 GHz Intel R Core TM i7-6700 processor and with 16GB RAM. |
| Software Dependencies | Yes | Just as in Fast LAS, this final solving stage can be encoded in ASP3 and solved using the ASP solver Clingo [Gebser et al., 2016]. |
| Experiment Setup | Yes | Full details of the flags used can be found in the supplementary material document. |