pith. sign in

arxiv: 2506.06739 · v3 · pith:2XUZK46Inew · submitted 2025-06-07 · 💻 cs.AI · cs.LG

Honey, I shrunk the hypothesis space (through logical preprocessing)

Andrew Cropper , Filipe Gouveia , David M. Cerna This is my paper

Pith reviewed 2026-05-22 00:49 UTC · model grok-4.3

classification 💻 cs.AI cs.LG
keywords inductive logic programminghypothesis space reductionlogical preprocessinganswer set programmingbackground knowledgemachine learningconstraint-based learning
0
0 comments X

The pith

A logical preprocessing step shrinks the hypothesis space for inductive logic programming by removing rules that cannot belong to any optimal hypothesis.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Inductive logic programming searches a hypothesis space of possible logical rules to find one that fits training examples and background knowledge. The paper shows how to use the background knowledge itself to deduce and discard rules that can never appear in an optimal hypothesis, regardless of what the examples are. Examples include determining that even numbers cannot be odd or that primes greater than 2 must be odd. The method is implemented with answer set programming and tested on a constraint-based ILP system. Experiments on visual reasoning and game playing show that ten seconds of preprocessing can reduce learning times from over ten hours to two seconds while preserving predictive accuracy.

Core claim

The approach encodes background knowledge as logical constraints and uses answer set programming to identify rules that violate those constraints, then removes the violating rules from the hypothesis space before the ILP system begins its search.

What carries the argument

A preprocessing procedure in answer set programming that deduces forbidden rules from background knowledge independent of any training examples.

If this is right

  • ILP learning times can drop from hours to seconds after only brief preprocessing.
  • Predictive accuracy remains unchanged because optimal hypotheses are preserved.
  • The technique applies directly to domains such as visual reasoning and game playing.
  • It integrates with existing constraint-based ILP systems without altering their core search procedure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same logical preprocessing idea could be adapted to prune search spaces in other inductive or constraint-based learning systems.
  • Upfront investment in deduction from background knowledge may become a practical way to scale ILP to problems with very large hypothesis spaces.
  • This separation of logical filtering from data-driven search suggests a general pattern for making exhaustive search methods more efficient.

Load-bearing premise

Background knowledge must suffice to deduce rules that cannot be part of an optimal hypothesis for any possible training examples, and removing those rules must leave at least one optimal hypothesis intact.

What would settle it

A dataset in which a rule removed during preprocessing turns out to be necessary for any hypothesis that correctly explains the training examples and achieves optimal performance.

Figures

Figures reproduced from arXiv: 2506.06739 by Andrew Cropper, David M. Cerna, Filipe Gouveia.

Figure 1
Figure 1. Figure 1: Learning time improvements when using shrinker. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. iggp-duikoshi-next_control task, shrinker reduces the learning time from 6.5 ± 1 hours to 1 ± 0 seconds. For the iggp-horseshoe-terminal task, shrinker reduces the learning time from 10 ± 0 hours … view at source ↗
Figure 2
Figure 2. Figure 2: Learning time improvements when using shrinker with only unsatisfiable rules. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. -15 0 15 30 45 60 Task Learning time difference (minutes) -15 0 15 30 45 60 Task Learning time difference (minutes) [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Learning time improvements when using shrinker with only implication reducible rules. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. Limitations Finite BK. Our shrinking idea is sufficiently general to handle definite programs as BK. However, because our bottom-up implementation uses ASP, w… view at source ↗
Figure 4
Figure 4. Figure 4: Learning time improvements when using shrinker with only recall reducible rules. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. -15 0 15 30 45 60 Task Learning time difference (minutes) -15 0 15 30 45 60 Task Learning time difference (minutes) [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Learning time improvements when using shrinker with only singletons reducible rules. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. this limitation only applies if our approach is used with a system that does not make the CWA, such as recent rule mining algorithms [69]. Noisy BK. We assume … view at source ↗
Figure 6
Figure 6. Figure 6: Learning time improvements when using shrinker with 100 vs 10 second timeout. The left figure shows all tasks and the right figure shows tasks where the two approaches significantly (𝑝 < 0.05) differ. The tasks are ordered by the improvement. Acknowledgments Andrew Cropper and Filipe Gouveia were supported by the EPSRC fellowship (EP/V040340/1). David M. Cerna was supported by the Czech Science Foundation … view at source ↗
Figure 7
Figure 7. Figure 7: The relationship between implication and recall reducible with respect to the rules introduced in Example [PITH_FULL_IMAGE:figures/full_fig_p031_7.png] view at source ↗
read the original abstract

Inductive logic programming (ILP) is a form of logical machine learning. The goal is to search a hypothesis space for a hypothesis that generalises training examples and background knowledge. We introduce an approach that 'shrinks' the hypothesis space before an ILP system searches it. Our approach uses background knowledge to find rules that cannot be in an optimal hypothesis regardless of the training examples. For instance, our approach discovers relationships such as "even numbers cannot be odd" and "prime numbers greater than 2 are odd". It then removes violating rules from the hypothesis space. We implement our approach using answer set programming and use it to shrink the hypothesis space of a constraint-based ILP system. Our experiments on multiple domains, including visual reasoning and game playing, show that our approach can substantially reduce learning times whilst maintaining predictive accuracies. For instance, given just 10 seconds of preprocessing time, our approach can reduce learning times from over 10 hours to only 2 seconds.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a preprocessing technique for constraint-based inductive logic programming (ILP) that uses background knowledge (BK) and answer set programming (ASP) to deduce and remove rules from the hypothesis space that cannot appear in any optimal hypothesis, independent of the training examples. Examples include deducing that even numbers cannot be odd or that primes greater than 2 are odd, then pruning violating rules. Experiments on visual reasoning and game-playing domains report large reductions in learning time (e.g., from over 10 hours to 2 seconds after 10 seconds of preprocessing) while preserving predictive accuracy.

Significance. If the logical pruning step is sound and preserves at least one optimal solution, the method offers a principled way to reduce the effective size of the hypothesis space in ILP without introducing example-dependent bias. The independence from training data and use of existing BK are strengths that could improve scalability for systems that already encode domain constraints in ASP.

major comments (2)
  1. [Approach / Method] The central guarantee—that rules violating BK-derived properties (e.g., parity or primality) cannot belong to any optimal hypothesis—requires an explicit theorem or lemma showing that the ASP encoding removes only rules that are inconsistent with the BK while leaving at least one optimal solution intact. Without this, the claim that accuracy is maintained rests on the experimental outcomes alone.
  2. [Experiments] Experiments: the reported reduction from >10 hours to 2 seconds after 10 s preprocessing is a strong result, but the manuscript must specify the exact ILP solver, the size of the original hypothesis space before and after pruning, the number of runs, and confirmation that the recovered hypothesis is optimal (or at least as accurate as the unpruned baseline).
minor comments (2)
  1. [Implementation] Clarify the precise interface between the preprocessing ASP program and the downstream constraint-based ILP system (e.g., how pruned rules are removed from the mode declarations or constraint set).
  2. [Discussion] Add a short discussion of failure cases: when the BK is insufficient to deduce any prunable rules, or when the preprocessing itself becomes a bottleneck.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments and the opportunity to clarify and strengthen our manuscript. We address each major comment below.

read point-by-point responses

  1. Referee: [Approach / Method] The central guarantee—that rules violating BK-derived properties (e.g., parity or primality) cannot belong to any optimal hypothesis—requires an explicit theorem or lemma showing that the ASP encoding removes only rules that are inconsistent with the BK while leaving at least one optimal solution intact. Without this, the claim that accuracy is maintained rests on the experimental outcomes alone.

    Authors: We agree that an explicit formal statement would improve the presentation. In the revised manuscript we will insert a short lemma establishing that the ASP preprocessing removes precisely those rules that are inconsistent with the background knowledge (hence cannot appear in any hypothesis consistent with the BK) and that the procedure is guaranteed to retain at least one optimal hypothesis because it never deletes rules solely on the basis of example-dependent information. The argument relies on the monotonicity of answer-set semantics and the fact that optimality is defined with respect to the full hypothesis space prior to pruning. revision: yes

  2. Referee: [Experiments] Experiments: the reported reduction from >10 hours to 2 seconds after 10 s preprocessing is a strong result, but the manuscript must specify the exact ILP solver, the size of the original hypothesis space before and after pruning, the number of runs, and confirmation that the recovered hypothesis is optimal (or at least as accurate as the unpruned baseline).

    Authors: We will expand the experimental section to name the precise ILP solver, tabulate the hypothesis-space cardinalities before and after pruning for every domain, state the number of independent runs, and report a side-by-side accuracy comparison demonstrating that the pruned-space hypotheses achieve the same predictive accuracy as those obtained from the unpruned baseline. These additions will confirm that at least one optimal (accuracy-maximising) hypothesis is preserved. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper's core derivation uses answer set programming to extract logical constraints (e.g., parity or primality) directly from background knowledge and removes rules that violate them. Because any optimal hypothesis must remain consistent with the BK, the pruning step is a sound, example-independent deduction rather than a fit, self-definition, or self-citation reduction. The reported runtime gains follow from this logical preprocessing without altering the set of optimal solutions, and the argument is self-contained against external logical benchmarks with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests primarily on the domain assumption that background knowledge can identify rules that are never optimal independent of training data; no free parameters or invented entities are evident from the abstract.

axioms (1)
  • domain assumption Background knowledge can be used to deduce rules that cannot be in an optimal hypothesis regardless of the training examples.
    This premise directly justifies the removal of violating rules during preprocessing.

pith-pipeline@v0.9.0 · 5700 in / 1355 out tokens · 74931 ms · 2026-05-22T00:49:03.386631+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

71 extracted references · 71 canonical work pages · 1 internal anchor

  1. [1]

    John Ahlgren and Shiu Yin Yuen. 2013. Efficient program synthesis using constraint satisfaction in inductive logic programming. J. Machine Learning Res. 14, 1 (2013), 3649–3682

    work page 2013

  2. [2]

    Aaron Bembenek, Michael Greenberg, and Stephen Chong. 2023. From SMT to ASP: Solver-Based Approaches to Solving Datalog Synthesis-as-Rule-Selection Problems. Proc. ACM Program. Lang. 7, POPL, Article 7 (jan 2023), 33 pages. https://doi.org/10.1145/3571200

    work page doi:10.1145/3571200 2023

  3. [3]

    Hendrik Blockeel and Luc De Raedt. 1998. Top-Down Induction of First-Order Logical Decision Trees. Artif. Intell. 101, 1-2 (1998), 285–297

    work page 1998

  4. [4]

    Anselm Blumer, Andrzej Ehrenfeucht, David Haussler, and Manfred K. Warmuth. 1987. Occam’s Razor.Inf. Process. Lett. 24, 6 (1987), 377–380

    work page 1987

  5. [5]

    Will Bridewell and Ljupčo Todorovski. 2007. Learning declarative bias. In International Conference on Inductive Logic Programming . Springer, 63–77

    work page 2007

  6. [6]

    François Chollet. 2019. On the Measure of Intelligence. CoRR (2019)

    work page 2019

  7. [7]

    Domenico Corapi, Alessandra Russo, and Emil Lupu. 2011. Inductive Logic Programming in Answer Set Programming. In ILP 2011

    work page 2011

  8. [8]

    Andrew Cropper. 2019. Playgol: Learning Programs Through Play. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019, Macao, China, August 10-16, 2019 . 6074–6080

    work page 2019

  9. [9]

    Andrew Cropper and Sebastijan Dumancic. 2022. Inductive Logic Programming At 30: A New Introduction. J. Artif. Intell. Res. 74 (2022), 765–850. https://doi.org/10.1613/jair.1.13507

    work page doi:10.1613/jair.1.13507 2022

  10. [10]

    Andrew Cropper, Richard Evans, and Mark Law. 2020. Inductive general game playing. Mach. Learn. 109, 7 (2020), 1393–1434. https://doi.org/10.1007/s10994-019-05843-w

    work page doi:10.1007/s10994-019-05843-w 2020

  11. [11]

    Andrew Cropper and Céline Hocquette. 2023. Learning Logic Programs by Combining Programs. InECAI 2023 - 26th European Conference on Artificial Intelligence, Vol. 372. IOS Press, 501–508. https://doi.org/10.3233/FAIA230309

    work page doi:10.3233/faia230309 2023

  12. [12]

    Andrew Cropper and Céline Hocquette. 2023. Learning Logic Programs by Discovering Where Not to Search. In Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023 . AAAI Press, 6289–6296. https://doi.org/10.1609/AAAI.V37I5.25774

    work page doi:10.1609/aaai.v37i5.25774 2023

  13. [13]

    Andrew Cropper and Rolf Morel. 2021. Learning programs by learning from failures. Mach. Learn. 110, 4 (2021), 801–856. https: //doi.org/10.1007/s10994-020-05934-z

    work page doi:10.1007/s10994-020-05934-z 2021

  14. [14]

    Andrew Cropper and Sophie Tourret. 2020. Logical reduction of metarules. Mach. Learn. 109, 7 (2020), 1323–1369. https://doi.org/10. 1007/s10994-019-05834-x JAIR, Vol. 4, Article 6. Publication date: August 2025. 6:26 • Andrew Cropper, Filipe Gouveia, and David M. Cerna

    work page 2020

  15. [15]

    Wang-Zhou Dai and Stephen Muggleton. 2021. Abductive Knowledge Induction from Raw Data. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI 2021, Virtual Event / Montreal, Canada, 19-27 August 2021 . 1845–1851. https://doi.org/10.24963/ijcai.2021/254

    work page doi:10.24963/ijcai.2021/254 2021

  16. [16]

    Evgeny Dantsin, Thomas Eiter, Georg Gottlob, and Andrei Voronkov. 2001. Complexity and expressive power of logic programming. ACM Comput. Surv. 33, 3 (2001), 374–425

    work page 2001

  17. [17]

    Luc De Raedt. 2008. Logical and relational learning . https://doi.org/10.1007/978-3-540-68856-3

    work page doi:10.1007/978-3-540-68856-3 2008

  18. [18]

    Sebastijan Dumancic, Tias Guns, and Andrew Cropper. 2021. Knowledge Refactoring for Inductive Program Synthesis. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021 . 7271–7278. https://ojs.aaai.org/index.php/AAAI/article/view/16893

    work page 2021

  19. [19]

    Sebastijan Dumančić, Tias Guns, Wannes Meert, and Hendrik Blockeel. 2019. Learning Relational Representations with Auto-encoding Logic Programs. In IJCAI 2019. 6081–6087

    work page 2019

  20. [20]

    Niklas Eén and Armin Biere. 2005. Effective Preprocessing in SAT Through Variable and Clause Elimination. InSAT 2005, Vol. 3569. 61–75. https://doi.org/10.1007/11499107_5

    work page doi:10.1007/11499107_5 2005

  21. [21]

    Kevin Ellis, Lucas Morales, Mathias Sablé-Meyer, Armando Solar-Lezama, and Josh Tenenbaum. 2018. Learning Libraries of Subroutines for Neurally-Guided Bayesian Program Induction. In NeurIPS 2018. 7816–7826

    work page 2018

  22. [22]

    Richard Evans and Edward Grefenstette. 2018. Learning Explanatory Rules from Noisy Data. J. Artif. Intell. Res. 61 (2018), 1–64

    work page 2018

  23. [23]

    Richard Evans, José Hernández-Orallo, Johannes Welbl, Pushmeet Kohli, and Marek J. Sergot. 2021. Making sense of sensory input. Artif. Intell. 293 (2021), 103438. https://doi.org/10.1016/j.artint.2020.103438

    work page doi:10.1016/j.artint.2020.103438 2021

  24. [24]

    Fonseca, Vítor Santos Costa, Fernando M

    Nuno A. Fonseca, Vítor Santos Costa, Fernando M. A. Silva, and Rui Camacho. 2004. On Avoiding Redundancy in Inductive Logic Programming. In Inductive Logic Programming, 14th International Conference, ILP 2004, Porto, Portugal, September 6-8, 2004, Proceedings (Lecture Notes in Computer Science, Vol. 3194) . Springer, 132–146. https://doi.org/10.1007/978-3...

    work page doi:10.1007/978-3-540-30109-7_13 2004

  25. [25]

    Johannes Fürnkranz and Tomás Kliegr. 2015. A Brief Overview of Rule Learning. In RuleML 2015 (Lecture Notes in Computer Science, Vol. 9202). Springer, 54–69. https://doi.org/10.1007/978-3-319-21542-6_4

    work page doi:10.1007/978-3-319-21542-6_4 2015

  26. [26]

    Suchanek

    Luis Galárraga, Christina Teflioudi, Katja Hose, and Fabian M. Suchanek. 2015. Fast rule mining in ontological knowledge bases with AMIE+. VLDB J. 24, 6 (2015), 707–730

    work page 2015

  27. [27]

    Martin Gebser, Roland Kaminski, Benjamin Kaufmann, and Torsten Schaub. 2012. Answer Set Solving in Practice

    work page 2012

  28. [28]

    Martin Gebser, Roland Kaminski, Benjamin Kaufmann, and Torsten Schaub. 2014. Clingo = ASP + Control: Preliminary Report. CoRR abs/1405.3694 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  29. [29]

    Martin Gebser, Roland Kaminski, Benjamin Kaufmann, and Torsten Schaub. 2019. Multi-shot ASP solving with clingo. Theory Pract. Log. Program. 19, 1 (2019), 27–82. https://doi.org/10.1017/S1471068418000054

    work page doi:10.1017/s1471068418000054 2019

  30. [30]

    Genesereth and Yngvi Björnsson

    Michael R. Genesereth and Yngvi Björnsson. 2013. The International General Game Playing Competition. AI Magazine 34, 2 (2013), 107–111. http://www.aaai.org/ojs/index.php/aimagazine/article/view/2475

    work page 2013

  31. [31]

    Fermüller

    Georg Gottlob and Christian G. Fermüller. 1993. Removing Redundancy from a Clause. Artif. Intell. 61, 2 (1993), 263–289. https: //doi.org/10.1016/0004-3702(93)90069-N

    work page doi:10.1016/0004-3702(93)90069-n 1993

  32. [32]

    Sumit Gulwani, Oleksandr Polozov, Rishabh Singh, et al. 2017. Program synthesis. Foundations and Trends® in Programming Languages 4, 1-2 (2017), 1–119

    work page 2017

  33. [33]

    Burghouts

    Fieke Hillerström and Gertjan J. Burghouts. 2024. Towards Probabilistic Inductive Logic Programming with Neurosymbolic Inference and Relaxation. Theory Pract. Log. Program. 24, 4 (2024), 628–643. https://doi.org/10.1017/S1471068424000371

    work page doi:10.1017/s1471068424000371 2024

  34. [34]

    Céline Hocquette and Andrew Cropper. 2024. Relational decomposition for program synthesis. CoRR abs/2408.12212 (2024). https: //doi.org/10.48550/ARXIV.2408.12212 arXiv:2408.12212

    work page doi:10.48550/arxiv.2408.12212 2024

  35. [35]

    Muggleton

    Céline Hocquette and Stephen H. Muggleton. 2020. Complete Bottom-Up Predicate Invention in Meta-Interpretive Learning. In IJCAI

    work page 2020

  36. [36]

    https://doi.org/10.24963/ijcai.2020/320

    2312–2318. https://doi.org/10.24963/ijcai.2020/320

    work page doi:10.24963/ijcai.2020/320 2020

  37. [37]

    Céline Hocquette, Andreas Niskanen, Matti Järvisalo, and Andrew Cropper. 2024. Learning MDL Logic Programs from Noisy Data. In Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI . AAAI Press, 10553–10561. https://doi.org/10.1609/AAAI.V38I9.28925

    work page doi:10.1609/aaai.v38i9.28925 2024

  38. [38]

    Céline Hocquette, Andreas Niskanen, Rolf Morel, Matti Järvisalo, and Andrew Cropper. 2024. Learning Big Logical Rules by Joining Small Rules. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI 2024, Jeju, South Korea, August 3-9, 2024. ijcai.org, 3430–3438. https://www.ijcai.org/proceedings/2024/380

    work page 2024

  39. [39]

    Krystof Hoder, Zurab Khasidashvili, Konstantin Korovin, and Andrei Voronkov. 2012. Preprocessing techniques for first-order clausifica- tion. In Formal Methods in Computer-Aided Design, FMCAD 2012, Cambridge, UK, October 22-25, 2012 , Gianpiero Cabodi and Satnam Singh (Eds.). IEEE, 44–51. https://ieeexplore.ieee.org/document/6462554/

    work page arXiv 2012

  40. [40]

    Pan Hu, Boris Motik, and Ian Horrocks. 2018. Optimised Maintenance of Datalog Materialisations. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), the 30th innovative Applications of Artificial Intelligence (IAAI-18), and the 8th AAAI Symposium on Educational Advances in Artificial Intelligence (EAAI-18), New Orlean...

    work page doi:10.1609/aaai.v32i1.11554 2018

  41. [41]

    Katsumi Inoue, Andrei Doncescu, and Hidetomo Nabeshima. 2013. Completing causal networks by meta-level abduction. Mach. Learn. 91, 2 (2013), 239–277. JAIR, Vol. 4, Article 6. Publication date: August 2025. Honey, I Shrunk the Hypothesis Space (Through Logical Preprocessing) • 6:27

    work page 2013

  42. [42]

    William Joyner. 1976. Resolution Strategies as Decision Procedures. J. ACM 23, 3 (1976), 398–417. https://doi.org/10.1145/321958.321960

    work page doi:10.1145/321958.321960 1976

  43. [43]

    Tobias Kaminski, Thomas Eiter, and Katsumi Inoue. 2019. Meta-Interpretive Learning Using HEX-Programs. In IJCAI 2019. 6186–6190. https://doi.org/10.24963/ijcai.2019/860

    work page doi:10.24963/ijcai.2019/860 2019

  44. [44]

    Zurab Khasidashvili and Konstantin Korovin. 2016. Predicate Elimination for Preprocessing in First-Order Theorem Proving. InTheory and Applications of Satisfiability Testing - SAT 2016 - 19th International Conference, Bordeaux, France, July 5-8, 2016, Proceedings (Lecture Notes in Computer Science, Vol. 9710), Nadia Creignou and Daniel Le Berre (Eds.). Sp...

    work page doi:10.1007/978-3-319-40970-2_22 2016

  45. [45]

    King, Michael J

    Ross D. King, Michael J. E. Sternberg, and Ashwin Srinivasan. 1995. Relating Chemical Activity to Structure: An Examination of ILP Successes. New Gener. Comput. 13, 3&4 (1995), 411–433. https://doi.org/10.1007/BF03037232

    work page doi:10.1007/bf03037232 1995

  46. [46]

    Oliver Kullmann. 1999. On a Generalization of Extended Resolution. Discret. Appl. Math. 96-97 (1999), 149–176. https://doi.org/10.1016/ S0166-218X(99)00037-2

    work page 1999

  47. [47]

    Larson and Ryszard S

    J. Larson and Ryszard S. Michalski. 1977. Inductive inference of VL decision rules. SIGART Newsletter 63 (1977), 38–44

    work page 1977

  48. [48]

    Mark Law, Alessandra Russo, and Krysia Broda. 2014. Inductive Learning of Answer Set Programs. In JELIA 2014

    work page 2014

  49. [49]

    John W Lloyd. 2012. Foundations of logic programming . Springer Science & Business Media

    work page 2012

  50. [50]

    Eric McCreath and Arun Sharma. 1995. Extraction of Meta-Knowledge to Restrict the Hypothesis Space for ILP Systems. In Eighth Australian Joint Conference on Artificial Intelligence . 75–82

    work page 1995

  51. [51]

    Huynh, and Raymond J

    Lilyana Mihalkova, Tuyen N. Huynh, and Raymond J. Mooney. 2007. Mapping and Revising Markov Logic Networks for Transfer Learning. In AAAI 2007. 608–614

    work page 2007

  52. [52]

    Stephen Muggleton. 1991. Inductive Logic Programming. New Generation Computing 8, 4 (1991), 295–318

    work page 1991

  53. [53]

    Stephen Muggleton. 1995. Inverse Entailment and Progol. New Generation Comput. 13, 3&4 (1995), 245–286

    work page 1995

  54. [54]

    Muggleton, Dianhuan Lin, and Alireza Tamaddoni-Nezhad

    Stephen H. Muggleton, Dianhuan Lin, and Alireza Tamaddoni-Nezhad. 2015. Meta-interpretive learning of higher-order dyadic Datalog: predicate invention revisited. Mach. Learn. 100, 1 (2015), 49–73

    work page 2015

  55. [55]

    Jose Picado, Arash Termehchy, Alan Fern, and Sudhanshu Pathak. 2017. Towards Automatically Setting Language Bias in Relational Learning. In Proceedings of the 1st Workshop on Data Management for End-to-End Machine Learning, DEEM@SIGMOD 2017, Chicago, IL, USA, May 14, 2017 . 3:1–3:4. https://doi.org/10.1145/3076246.3076249

    work page doi:10.1145/3076246.3076249 2017

  56. [56]

    G.D. Plotkin. 1971. Automatic Methods of Inductive Inference . Ph. D. Dissertation. Edinburgh University

    work page 1971

  57. [57]

    Purgal, David M

    Stanislaw J. Purgal, David M. Cerna, and Cezary Kaliszyk. 2022. Learning Higher-Order Logic Programs From Failures. In Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, IJCAI 2022, Vienna, Austria, 23-29 July 2022 , Luc De Raedt (Ed.). ijcai.org, 2726–2733. https://doi.org/10.24963/IJCAI.2022/378

    work page doi:10.24963/ijcai.2022/378 2022

  58. [58]

    Ross Quinlan

    J. Ross Quinlan. 1990. Learning Logical Definitions from Relations. Mach. Learn. 5 (1990), 239–266

    work page 1990

  59. [59]

    Luc De Raedt and Jan Ramon. 2004. Condensed Representations for Inductive Logic Programming. In Principles of Knowledge Repre- sentation and Reasoning: Proceedings of the Ninth International Conference (KR2004), Whistler, Canada, June 2-5, 2004 , Didier Dubois, Christopher A. Welty, and Mary-Anne Williams (Eds.). AAAI Press, 438–446. http://www.aaai.org/L...

    work page 2004

  60. [60]

    Oliver Ray. 2009. Nonmonotonic abductive inductive learning. J. Applied Logic 7, 3 (2009), 329–340

    work page 2009

  61. [61]

    Raymond Reiter. 1977. On Closed World Data Bases. In Logic and Data Bases, Symposium on Logic and Data Bases . 55–76

    work page 1977

  62. [62]

    Rule, Steven T

    Joshua S. Rule, Steven T. Piantadosi, Andrew Cropper, Kevin Ellis, Maxwell Nye, and Joshua B. Tenenbaum. 2024. Symbolic metaprogram search improves learning efficiency and explains rule learning in humans. Nature Communications 15, 1 (2024), 6847. https://doi.org/10. 1038/s41467-024-50966-x

    work page 2024

  63. [63]

    Iztok Savnik and Peter A Flach. 1993. Bottom-up induction of functional dependencies from relations. In Proceedings of the AAAI-93 Workshop on Knowledge Discovery in Databases . 174–185

    work page 1993

  64. [64]

    Peter Schüller and Mishal Benz. 2018. Best-effort inductive logic programming via fine-grained cost-based hypothesis generation - The inspire system at the inductive logic programming competition. Mach. Learn. 107, 7 (2018), 1141–1169. https://doi.org/10.1007/s10994- 018-5708-2

    work page doi:10.1007/s10994- 2018

  65. [65]

    Ashwin Srinivasan. 2001. The ALEPH manual. Machine Learning at the Computing Laboratory, Oxford University (2001)

    work page 2001

  66. [66]

    Ashwin Srinivasan and Ravi Kothari. 2005. A Study of Applying Dimensionality Reduction to Restrict the Size of a Hypothesis Space. In Inductive Logic Programming, 15th International Conference, ILP 2005, Bonn, Germany, August 10-13, 2005, Proceedings , Vol. 3625. 348–365

    work page 2005

  67. [67]

    Jan Struyf and Hendrik Blockeel. 2003. Query Optimization in Inductive Logic Programming by Reordering Literals. In Inductive Logic Programming: 13th International Conference, ILP 2003, Szeged, Hungary, September 29-October 1, 2003, Proceedings (Lecture Notes in Computer Science, Vol. 2835), Tamás Horváth (Ed.). Springer, 329–346. https://doi.org/10.1007/...

    work page doi:10.1007/978-3-540-39917-9_22 2003

  68. [68]

    Petar Vukmirovic, Jasmin Blanchette, and Marijn J. H. Heule. 2023. SAT-Inspired Eliminations for Superposition. ACM Trans. Comput. Log. 24, 1 (2023), 7:1–7:25. https://doi.org/10.1145/3565366

    work page doi:10.1145/3565366 2023

  69. [69]

    Yudong Xu, Wenhao Li, Pashootan Vaezipoor, Scott Sanner, and Elias Boutros Khalil. 2024. LLMs and the Abstraction and Reasoning Corpus: Successes, Failures, and the Importance of Object-based Representations. Trans. Mach. Learn. Res. 2024 (2024). https: //openreview.net/forum?id=E8m8oySvPJ

    work page 2024

  70. [70]

    Václav Zeman, Tomás Kliegr, and Vojtech Svátek. 2021. RDFRules: Making RDF rule mining easier and even more efficient. Semantic Web 12, 4 (2021), 569–602. https://doi.org/10.3233/SW-200413 JAIR, Vol. 4, Article 6. Publication date: August 2025. 6:28 • Andrew Cropper, Filipe Gouveia, and David M. Cerna

    work page doi:10.3233/sw-200413 2021

  71. [71]

    cherry-picked

    Qiang Zeng, Jignesh M. Patel, and David Page. 2014. QuickFOIL: Scalable Inductive Logic Programming. VLDB (2014), 197–208. Appendix A We assume a simple typing mechanism with the properties described below. Definition 15 (Types). Let 𝐵 be background knowledge. Then 𝑡𝑦𝑝𝑒𝑠 (𝐵) is a set unary predicate symbols defined in 𝐵 with the following properties: For ...

    work page 2014