pith. sign in

arxiv: 2509.21663 · v2 · pith:TMEOLZISnew · submitted 2025-09-25 · 💻 cs.LG · cs.AI· cs.LO

Logic of Hypotheses: from Zero to Full Knowledge in Neurosymbolic Integration

Davide Bizzaro , Alessandro Daniele This is my paper

Pith reviewed 2026-05-21 21:45 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.LO
keywords neurosymbolic integrationlogic of hypotheseschoice operatorGödel fuzzy logicdifferentiable compilationrule inductionsymbolic priorsdiscretization
0
0 comments X

The pith

Logic of Hypotheses unifies hand-crafted rules and data-induced rules in neurosymbolic models by adding a learnable choice operator to propositional logic.

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

The paper introduces Logic of Hypotheses to merge two strands of neurosymbolic work: one that injects fixed expert rules into neural networks and one that induces rules purely from data. It extends propositional logic with a choice operator whose parameters are learned during training. Formulas written in this language compile directly into differentiable graphs through fuzzy logic, so backpropagation selects the best options. The resulting models support any degree of prior knowledge and convert to exact Boolean functions with no performance loss when Gödel fuzzy logic and the Gödel trick are used. This gives a single framework that starts from zero knowledge or full knowledge and still produces interpretable logical programs.

Core claim

Logic of Hypotheses extends propositional logic with a choice operator that has learnable parameters and selects among subformulas; when interpreted with Gödel fuzzy logic the formulas become differentiable programs trainable by gradient descent, subsuming existing neurosymbolic models and allowing discretization to hard Boolean functions without accuracy loss via the Gödel trick.

What carries the argument

The choice operator, an extension to propositional logic syntax that selects one subformula from a pool according to learnable parameters optimized by backpropagation.

If this is right

  • Models can be built with any mixture of fixed symbolic knowledge and learned components.
  • Existing neurosymbolic systems appear as special cases where some or all choices are fixed in advance.
  • Trained models convert to exact logical programs that match the performance of the differentiable version.
  • The same framework handles both tabular classification and tasks that combine perception with reasoning.

Where Pith is reading between the lines

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

  • The choice operator could be added to other logical languages to increase their flexibility in hybrid systems.
  • Success of discretization on the reported tasks suggests checking whether the same zero-loss conversion holds on larger-scale reasoning benchmarks.
  • The unification may simplify transfer of trained models between different neurosymbolic implementations.

Load-bearing premise

Formulas in Logic of Hypotheses can be compiled into a differentiable graph using fuzzy logic so that choice parameters are learned via backpropagation, and the Gödel trick converts the trained model to a Boolean function with no performance loss.

What would settle it

A trained Logic of Hypotheses model that shows lower accuracy on held-out data after the Gödel trick is applied to produce a hard Boolean version than it showed in its original fuzzy form.

Figures

Figures reproduced from arXiv: 2509.21663 by Alessandro Daniele, Davide Bizzaro.

Figure 1
Figure 1. Figure 1: Given a choice operator [a, b, c], the figure illustrates the differentiable three-stage procedure that turns the raw, real-valued logits zi attached to each candidate sub-formula into logical gates wi ∈ [0, 1]. From the bottom up: (1) during training, i.i.d. Gumbel noise ni ∼ Gumbel(0, β) is added to each logit; (2) the mean z¯ ′ of the two largest perturbed logits is subtracted from all values; (3) tempe… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of disjunctive and conjunctive compilations. [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Average training curves — over 20 runs — of LoH formulas following different templates, [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Clauses selection performance w.r.t. percentage of true labels, number of ground-truth [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
read the original abstract

Neurosymbolic integration (NeSy) blends neural-network learning with symbolic reasoning. The field can be split between methods injecting hand-crafted rules into neural models, and methods inducing symbolic rules from data. We introduce Logic of Hypotheses (LoH), a novel language that unifies these strands, enabling the flexible integration of data-driven rule learning with symbolic priors and expert knowledge. LoH extends propositional logic syntax with a choice operator, which has learnable parameters and selects a subformula from a pool of options. Using fuzzy logic, formulas in LoH can be directly compiled into a differentiable computational graph, so the optimal choices can be learned via backpropagation. This framework subsumes some existing NeSy models, while adding the possibility of arbitrary degrees of knowledge specification. Moreover, the use of G\"odel fuzzy logic and the recently developed G\"odel trick yields models that can be discretized to hard Boolean-valued functions without any loss in performance. We provide experimental analysis on such models, showing strong results on tabular data and on two NeSy tasks with a perceptual component.

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 / 1 minor

Summary. The manuscript introduces Logic of Hypotheses (LoH), an extension of propositional logic that adds a choice operator with learnable parameters. LoH unifies hand-crafted rule injection and data-driven symbolic rule induction within neurosymbolic integration, subsumes certain existing NeSy models, and permits arbitrary degrees of knowledge specification. Formulas compile directly to differentiable computational graphs via fuzzy logic, allowing optimal choices to be learned by backpropagation. The central technical claim is that Gödel fuzzy logic combined with the Gödel trick produces models that discretize exactly to hard Boolean-valued functions with no performance loss. Experimental results are reported on tabular data and two perceptual NeSy tasks.

Significance. If the discretization property and unification hold with the claimed exactness, LoH would offer a flexible, principled bridge between symbolic priors and neural learning while enabling crisp, interpretable models. The ability to move from partial to full knowledge specification without performance degradation would be a useful contribution to the NeSy literature. The manuscript's experimental section is noted as showing strong results, but the absence of supporting derivations for the zero-loss claim limits the assessed impact at present.

major comments (2)
  1. [Abstract] Abstract, final paragraph: the claim that Gödel fuzzy logic plus the Gödel trick yields discretization to hard Boolean functions 'without any loss in performance' is presented without derivation, error bounds, or analysis of the choice-operator parameters. It is unclear whether back-propagation on the learnable parameters is guaranteed to reach configurations where fuzzy min/max semantics coincide exactly with Boolean evaluation on the data support, or whether gradient issues or local minima could produce non-exact discretizations.
  2. [Abstract] Abstract: no quantitative results, error analysis, or ablation on the discretization step are supplied despite the strong claim of 'strong results' and zero performance loss. This makes it impossible to verify whether the central discretization property is load-bearing or merely asserted.
minor comments (1)
  1. [Abstract] The 'Gödel trick' is referenced but not defined or referenced in the abstract; a short explanation or citation would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's detailed feedback on our manuscript. We address the concerns about the presentation of the discretization claim in the abstract and the lack of quantitative details there. The full paper provides the supporting derivations and experimental results, but we agree that the abstract could benefit from additional clarification and a brief summary of results.

read point-by-point responses

  1. Referee: [Abstract] Abstract, final paragraph: the claim that Gödel fuzzy logic plus the Gödel trick yields discretization to hard Boolean functions 'without any loss in performance' is presented without derivation, error bounds, or analysis of the choice-operator parameters. It is unclear whether back-propagation on the learnable parameters is guaranteed to reach configurations where fuzzy min/max semantics coincide exactly with Boolean evaluation on the data support, or whether gradient issues or local minima could produce non-exact discretizations.

    Authors: We thank the referee for highlighting this point. The derivation of the exact discretization property using Gödel fuzzy logic and the Gödel trick is detailed in Section 3 of the manuscript, where we show that the fuzzy min/max operations coincide with Boolean AND/OR when the choice parameters are optimized to select the appropriate subformulas, with no performance loss on the data support. Regarding the analysis of choice-operator parameters, we provide bounds and show that the parameters can be learned to achieve exact Boolean behavior. While we do not provide a formal proof that backpropagation is guaranteed to avoid all local minima in every possible case (as is common in optimization of neural networks), our theoretical analysis and empirical results indicate that the optimization landscape allows reaching the exact discretization configurations. We will revise the abstract to include a brief reference to this section and a short note on the parameter analysis. revision: partial

  2. Referee: [Abstract] Abstract: no quantitative results, error analysis, or ablation on the discretization step are supplied despite the strong claim of 'strong results' and zero performance loss. This makes it impossible to verify whether the central discretization property is load-bearing or merely asserted.

    Authors: We acknowledge that the abstract does not contain specific quantitative results or ablations, which is typical due to length constraints. The manuscript's experimental section reports results on tabular data and two perceptual NeSy tasks, including ablations that demonstrate the discretization maintains performance with zero loss. Error analysis is provided in the relevant sections. To address this, we will update the abstract to include a high-level summary of the key experimental findings supporting the zero-loss claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity in LoH derivation chain.

full rationale

The paper defines LoH as an extension of propositional logic with a choice operator having learnable parameters. Formulas are compiled to differentiable graphs via fuzzy logic for backpropagation-based learning of choices. The discretization property to hard Boolean functions with no performance loss is presented as a consequence of using Gödel fuzzy logic together with the recently developed Gödel trick. No quoted equation or step in the abstract or context reduces a claimed prediction or result to a fitted input by construction, nor does any central claim rest on a load-bearing self-citation chain that itself lacks independent verification. The unification of hand-crafted and data-induced rule methods, subsumption of existing NeSy models, and experimental results on tabular and perceptual tasks supply independent content outside any definitional loop.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the new choice operator, the assumption that fuzzy logic yields a differentiable graph, and the Gödel trick for lossless discretization; learnable parameters of the choice operator are the main free elements introduced.

free parameters (1)
  • learnable parameters of the choice operator
    Parameters that select among sub-formula options and are optimized via backpropagation.
axioms (1)
  • domain assumption Fuzzy logic semantics allow direct compilation of LoH formulas into a differentiable computational graph
    Invoked to enable gradient-based learning of the choice parameters.
invented entities (1)
  • Choice operator no independent evidence
    purpose: Extends propositional logic syntax to select a subformula from a pool with learnable parameters
    New syntactic construct introduced by the paper; no independent evidence supplied in the abstract.

pith-pipeline@v0.9.0 · 5718 in / 1364 out tokens · 59500 ms · 2026-05-21T21:45:51.776856+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

56 extracted references · 56 canonical work pages · 4 internal anchors

  1. [1]

    UCI Machine Learning Repository, 2012

    BLOGGER . UCI Machine Learning Repository, 2012. DOI : https://doi.org/10.24432/C5HK6P

    work page doi:10.24432/c5hk6p 2012

  2. [2]

    Tic-Tac-Toe Endgame

    David Aha. Tic-Tac-Toe Endgame . UCI Machine Learning Repository, 1991

    work page 1991

  3. [3]

    Logic tensor networks

    Samy Badreddine, Artur d'Avila Garcez, Luciano Serafini, and Michael Spranger. Logic tensor networks. Artificial Intelligence, 303: 0 103649, 2022

    work page 2022

  4. [4]

    Chess (King-Rook vs

    Michael Bain and Arthur Hoff. Chess (King-Rook vs. King) . UCI Machine Learning Repository, 1994. DOI : https://doi.org/10.24432/C57W2S

    work page doi:10.24432/c57w2s 1994

  5. [5]

    Interpretable neural-symbolic concept reasoning

    Pietro Barbiero, Gabriele Ciravegna, Francesco Giannini, Mateo Espinosa Zarlenga, Lucie Charlotte Magister, Alberto Tonda, Pietro Lio, Frederic Precioso, Mateja Jamnik, and Giuseppe Marra. Interpretable neural-symbolic concept reasoning. In Proceedings of the 40th International Conference on Machine Learning, volume 202, pp.\ 1801--1825, 2023

    work page 2023

  6. [6]

    Barry Becker and Ronny Kohavi. Adult . UCI Machine Learning Repository, 1996. DOI : https://doi.org/10.24432/C5XW20

    work page doi:10.24432/c5xw20 1996

  7. [7]

    Algorithms for hyper-parameter optimization

    James Bergstra, R\' e mi Bardenet, Yoshua Bengio, and Bal\' a zs K\' e gl. Algorithms for hyper-parameter optimization. In Advances in Neural Information Processing Systems, volume 24. Curran Associates, Inc., 2011. URL https://proceedings.neurips.cc/paper_files/paper/2011/file/86e8f7ab32cfd12577bc2619bc635690-Paper.pdf

    work page 2011

  8. [8]

    Neural-symbolic learning and reasoning: A survey and interpretation

    Tarek R Besold, Sebastian Bader, Howard Bowman, Pedro Domingos, Pascal Hitzler, Kai-Uwe K \"u hnberger, Luis C Lamb, Priscila Machado Vieira Lima, Leo de Penning, Gadi Pinkas, et al. Neural-symbolic learning and reasoning: A survey and interpretation. In Neuro-Symbolic Artificial Intelligence: The State of the Art, pp.\ 1--51. IOS press, 2021

    work page 2021

  9. [9]

    R. Bock. MAGIC Gamma Telescope . UCI Machine Learning Repository, 2004. DOI : https://doi.org/10.24432/C52C8B

    work page doi:10.24432/c52c8b 2004

  10. [10]

    Logical Rule Induction and Theory Learning Using Neural Theorem Proving

    Andres Campero, Aldo Pareja, Tim Klinger, Josh Tenenbaum, and Sebastian Riedel. Logical rule induction and theory learning using neural theorem proving. arXiv preprint arXiv:1809.02193, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  11. [11]

    Assessing satnet's ability to solve the symbol grounding problem

    Oscar Chang, Lampros Flokas, Hod Lipson, and Michael Spranger. Assessing satnet's ability to solve the symbol grounding problem. Advances in Neural Information Processing Systems, 33: 0 1428--1439, 2020

    work page 2020

  12. [12]

    Cerdeira, F

    Paulo Cortez, A. Cerdeira, F. Almeida, T. Matos, and J. Reis. Wine Quality . UCI Machine Learning Repository, 2009. DOI : https://doi.org/10.24432/C56S3T

    work page doi:10.24432/c56s3t 2009

  13. [13]

    Bridging machine learning and logical reasoning by abductive learning

    Wang-Zhou Dai, Qiuling Xu, Yang Yu, and Zhi-Hua Zhou. Bridging machine learning and logical reasoning by abductive learning. Advances in Neural Information Processing Systems, 32, 2019

    work page 2019

  14. [14]

    Knowledge enhanced neural networks

    Alessandro Daniele and Luciano Serafini. Knowledge enhanced neural networks. In Proceedings of the 16th Pacific Rim International Conference on Artificial Intelligence ( PRICAI ) , Lecture Notes in Computer Science, pp.\ 542--554. Springer, 2019

    work page 2019

  15. [15]

    Gradient-Based Optimization on G\"odel Logic as Discrete Local Search

    Alessandro Daniele and Emile van Krieken. Noise to the rescue: Escaping local minima in neurosymbolic local search, 2025. URL https://arxiv.org/abs/2503.01817

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  16. [16]

    Deep symbolic learning: Discovering symbols and rules from perceptions

    Alessandro Daniele, Tommaso Campari, Sagar Malhotra, and Luciano Serafini. Deep symbolic learning: Discovering symbols and rules from perceptions. arXiv preprint arXiv:2208.11561, 2022

    work page arXiv 2022

  17. [17]

    Gxnor-net: Training deep neural networks with ternary weights and activations without full-precision memory under a unified discretization framework

    Lei Deng, Peng Jiao, Jing Pei, Zhenzhi Wu, and Guoqi Li. Gxnor-net: Training deep neural networks with ternary weights and activations without full-precision memory under a unified discretization framework. Neural Networks, 100: 0 49--58, 2018

    work page 2018

  18. [18]

    Rl-net: Interpretable rule learning with neural networks

    Lucile Dierckx, Rosana Veroneze, and Siegfried Nijssen. Rl-net: Interpretable rule learning with neural networks. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp.\ 95--107. Springer, 2023

    work page 2023

  19. [19]

    Semantic-based regularization for learning and inference

    Michelangelo Diligenti, Marco Gori, and Claudio Sacca. Semantic-based regularization for learning and inference. Artificial Intelligence, 244: 0 143--165, 2017

    work page 2017

  20. [20]

    Supervised and unsupervised discretization of continuous features

    James Dougherty, Ron Kohavi, and Mehran Sahami. Supervised and unsupervised discretization of continuous features. ICML, 1995, 09 1997. doi:10.1016/B978-1-55860-377-6.50032-3

    work page doi:10.1016/b978-1-55860-377-6.50032-3 1995

  21. [21]

    Learning explanatory rules from noisy data

    Richard Evans and Edward Grefenstette. Learning explanatory rules from noisy data. Journal of Artificial Intelligence Research, 61: 0 1--64, 2018

    work page 2018

  22. [22]

    An algorithm for quadratic programming

    Marguerite Frank, Philip Wolfe, et al. An algorithm for quadratic programming. Naval research logistics quarterly, 3 0 (1-2): 0 95--110, 1956

    work page 1956

  23. [23]

    Differentiable rule induction from raw sequence inputs

    Kun Gao, Katsumi Inoue, Yongzhi Cao, Hanpin Wang, and Yang Feng. Differentiable rule induction from raw sequence inputs. In The Thirteenth International Conference on Learning Representations, 2025

    work page 2025

  24. [24]

    Statistical theory of extreme values and some practical applications: a series of lectures, volume 33

    Emil Julius Gumbel. Statistical theory of extreme values and some practical applications: a series of lectures, volume 33. US Government Printing Office, 1954

    work page 1954

  25. [25]

    Fast abductive learning by similarity-based consistency optimization

    Yu-Xuan Huang, Wang-Zhou Dai, Le-Wen Cai, Stephen H Muggleton, and Yuan Jiang. Fast abductive learning by similarity-based consistency optimization. Advances in Neural Information Processing Systems, 34: 0 26574--26584, 2021

    work page 2021

  26. [26]

    Categorical reparameterization with gumbel-softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. In International Conference on Learning Representations, 2017

    work page 2017

  27. [27]

    Anfis: adaptive-network-based fuzzy inference system

    J-SR Jang. Anfis: adaptive-network-based fuzzy inference system. IEEE transactions on systems, man, and cybernetics, 23 0 (3): 0 665--685, 1993

    work page 1993

  28. [28]

    Net-dnf: Effective deep modeling of tabular data

    Liran Katzir, Gal Elidan, and Ran El-Yaniv. Net-dnf: Effective deep modeling of tabular data. In International conference on learning representations, 2020

    work page 2020

  29. [29]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  30. [30]

    Differentiable rule induction with learned relational features

    Remy Kusters, Yusik Kim, Marine Collery, Christian de Sainte Marie, and Shubham Gupta. Differentiable rule induction with learned relational features. arXiv preprint arXiv:2201.06515, 2022

    work page arXiv 2022

  31. [31]

    Banknote Authentication

    Volker Lohweg. Banknote Authentication . UCI Machine Learning Repository, 2012. DOI : https://doi.org/10.24432/C55P57

    work page doi:10.24432/c55p57 2012

  32. [32]

    Deepproblog: Neural probabilistic logic programming

    Robin Manhaeve, Sebastijan Dumancic, Angelika Kimmig, Thomas Demeester, and Luc De Raedt. Deepproblog: Neural probabilistic logic programming. Advances in neural information processing systems, 31, 2018

    work page 2018

  33. [33]

    From statistical relational to neurosymbolic artificial intelligence: A survey

    Giuseppe Marra, Sebastijan Dumančić, Robin Manhaeve, and Luc De Raedt . From statistical relational to neurosymbolic artificial intelligence: A survey. Artificial Intelligence, 328: 0 104062, 2024. ISSN 0004-3702. doi:https://doi.org/10.1016/j.artint.2023.104062

    work page doi:10.1016/j.artint.2023.104062 2024

  34. [34]

    S. Moro, P. Rita, and P. Cortez. Bank Marketing . UCI Machine Learning Repository, 2014. DOI : https://doi.org/10.24432/C5K306

    work page doi:10.24432/c5k306 2014

  35. [35]

    Gsetsk: a generic self-evolving tsk fuzzy neural network with a novel hebbian-based rule reduction approach

    Ngoc Nam Nguyen, Weigui Jair Zhou, and Chai Quek. Gsetsk: a generic self-evolving tsk fuzzy neural network with a novel hebbian-based rule reduction approach. Applied Soft Computing, 35: 0 29--42, 2015. ISSN 1568-4946. doi:https://doi.org/10.1016/j.asoc.2015.06.008

    work page doi:10.1016/j.asoc.2015.06.008 2015

  36. [36]

    Inductive Logic Programming via Differentiable Deep Neural Logic Networks

    Ali Payani and Faramarz Fekri. Inductive logic programming via differentiable deep neural logic networks, 2019. URL https://arxiv.org/abs/1906.03523

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  37. [37]

    Deep differentiable logic gate networks

    Felix Petersen, Christian Borgelt, Hilde Kuehne, and Oliver Deussen. Deep differentiable logic gate networks. Advances in Neural Information Processing Systems, 35: 0 2006--2018, 2022

    work page 2006

  38. [38]

    Convolutional differentiable logic gate networks

    Felix Petersen, Hilde Kuehne, Christian Borgelt, Julian Welzel, and Stefano Ermon. Convolutional differentiable logic gate networks. Advances in Neural Information Processing Systems, 37: 0 121185--121203, 2024

    work page 2024

  39. [39]

    Learning accurate and interpretable decision rule sets from neural networks

    Litao Qiao, Weijia Wang, and Bill Lin. Learning accurate and interpretable decision rule sets from neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pp.\ 4303--4311, 2021

    work page 2021

  40. [40]

    Vladislav Rajkovic. Nursery . UCI Machine Learning Repository, 1989. DOI : https://doi.org/10.24432/C5P88W

    work page doi:10.24432/c5p88w 1989

  41. [41]

    Gray, Francois P

    Ryan Riegel, Alexander Gray, Francois Luus, Naweed Khan, Ndivhuwo Makondo, Ismail Yunus Akhalwaya, Haifeng Qian, Ronald Fagin, Francisco Barahona, Udit Sharma, et al. Logical neural networks. CoRR, abs/2006.13155, 2020. URL https://arxiv.org/abs/2006.13155

    work page arXiv 2006

  42. [42]

    Schlimmer

    J. Schlimmer. Mushroom . UCI Machine Learning Repository, 1981. DOI : https://doi.org/10.24432/C5959T

    work page doi:10.24432/c5959t 1981

  43. [43]

    Shihabudheen and G.N

    K.V. Shihabudheen and G.N. Pillai. Recent advances in neuro-fuzzy system: A survey. Knowledge-Based Systems, 152: 0 136--162, 2018. ISSN 0950-7051. doi:https://doi.org/10.1016/j.knosys.2018.04.014

    work page doi:10.1016/j.knosys.2018.04.014 2018

  44. [44]

    Letter Recognition

    David Slate. Letter Recognition . UCI Machine Learning Repository, 1991. DOI : https://doi.org/10.24432/C5ZP40

    work page doi:10.24432/c5zp40 1991

  45. [45]

    Techniques for symbol grounding with satnet

    Sever Topan, David Rolnick, and Xujie Si. Techniques for symbol grounding with satnet. Advances in Neural Information Processing Systems, 34: 0 20733--20744, 2021

    work page 2021

  46. [46]

    Connect-4

    John Tromp. Connect-4 . UCI Machine Learning Repository, 1995. DOI : https://doi.org/10.24432/C59P43

    work page doi:10.24432/c59p43 1995

  47. [47]

    Neural-symbolic integration: A compositional perspective

    Efthymia Tsamoura, Timothy Hospedales, and Loizos Michael. Neural-symbolic integration: A compositional perspective. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pp.\ 5051--5060, 2021

    work page 2021

  48. [48]

    SATN et: Bridging deep learning and logical reasoning using a differentiable satisfiability solver

    Po-Wei Wang, Priya Donti, Bryan Wilder, and Zico Kolter. SATN et: Bridging deep learning and logical reasoning using a differentiable satisfiability solver. In Proceedings of the 36th International Conference on Machine Learning, volume 97, pp.\ 6545--6554, 2019

    work page 2019

  49. [49]

    Transparent classification with multilayer logical perceptrons and random binarization

    Zhuo Wang, Wei Zhang, Jianyong Wang, et al. Transparent classification with multilayer logical perceptrons and random binarization. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pp.\ 6331--6339, 2020

    work page 2020

  50. [50]

    Deepstochlog: Neural stochastic logic programming

    Thomas Winters, Giuseppe Marra, Robin Manhaeve, and Luc De Raedt. Deepstochlog: Neural stochastic logic programming. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp.\ 10090--10100, 2022

    work page 2022

  51. [51]

    A semantic loss function for deep learning with symbolic knowledge

    Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang, and Guy Van den Broeck. A semantic loss function for deep learning with symbolic knowledge. In Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pp.\ 5502--5511. PMLR, 07 2018

    work page 2018

  52. [52]

    Neurasp: Embracing neural networks into answer set programming

    Zhun Yang, Adam Ishay, and Joohyung Lee. Neurasp: Embracing neural networks into answer set programming. In 29th International Joint Conference on Artificial Intelligence, IJCAI 2020, pp.\ 1755--1762. International Joint Conferences on Artificial Intelligence, 2020

    work page 2020

  53. [53]

    Zhou and C

    R.W. Zhou and C. Quek. A pseudo outer-product based fuzzy neural network and its rule-identification algorithm. In Proceedings of International Conference on Neural Networks (ICNN'96), pp.\ 1156--1161 vol.2, 1996. doi:10.1109/ICNN.1996.549061

    work page doi:10.1109/icnn.1996.549061 1996

  54. [54]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  55. [55]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  56. [56]

    o k w v0 g

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page arXiv