Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Reasoning in Neurosymbolic AI

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Every propositional logic formula can be built into a Restricted Boltzmann Machine whose minimum-energy states are exactly its satisfying assignments, turning logical reasoning into energy minimization.

desk verdict The SDNF-to-RBM construction is sound but is prior work; the new CNF translation (Eq. 18) is wrong and invalidates the SAT/MaxSAT experiments. read the letter →

arxiv 2505.20313 v1 pith:JLJFVLXQ submitted 2025-05-22 cs.AI cs.LO

Son Tran , Edjard Mota , Artur d'Avila Garcez This is my paper · ORCID
classification cs.AIcs.LO
keywords NeurosymbolicAIRestrictedBoltzmannMachinesLogicalpropositionallogicSATsolvingMaxenergy-basedlearningreasoning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that any propositional logic formula can be translated into a Restricted Boltzmann Machine (RBM) so that the formula's satisfying truth assignments are exactly the network's global energy minima. The resulting system, called the Logical Boltzmann Machine (LBM), therefore performs logical reasoning by searching for low-energy states. If correct, this gives a provably sound way to embed logical knowledge into a neural network and to add a verifiable reasoning module on top of deep networks that must respect constraints such as fairness or safety. The paper backs the claim with a proof, illustrative examples, and experiments on reasoning coverage, SAT and MaxSAT solving, and learning from data plus knowledge.

What carries the argument

The central object is the per-clause energy term $e_j(x,h_j) = -h_j(\sum_{t\in ST_j} x_t - \sum_{k\in SK_j} x_k - |ST_j| + \varepsilon)$. When the clause's literals are exactly satisfied, the expression inside the parentheses equals $\varepsilon$, so $e_j$ is minimized at $-\varepsilon$ with $h_j = 1$; otherwise it is non-negative and minimized at $0$ with $h_j = 0$. Summing these terms over all clauses makes the total energy count how many clauses are violated, so finding a satisfying assignment is equivalent to finding a global minimum of the RBM's energy (or, with confidence values, its free energy). This construction is what lets reasoning be carried out by Gibbs sampling, gradient descent, or global optimization methods.

What would settle it

Take a satisfiable random CNF with about 60 variables, compile it to an LBM, and run the paper's Gibbs-sampling procedure; if the sampler never reaches an energy minimum corresponding to a satisfying assignment within a fixed large number of steps while a symbolic solver confirms satisfiability, the claimed equivalence between sampling and reasoning fails for that instance class.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1: any strict DNF (SDNF) formula $\varphi \equiv \bigvee_j (\bigwedge_{t\in ST_j} x_t \wedge \bigwedge_{k\in SK_j} \neg x_k)$ can be mapped onto an RBM with energy function $E(x,h) = -\sum_j h_j(\sum_{t\in ST_j} x_t - \sum_{k\in SK_j} x_k - |ST_j| + \varepsilon)$, where $0<\varepsilon<1$, such that $s_\varphi(x) = -E(x)$. Because every well-formed formula can be converted into a full DNF and hence an SDNF, every propositional formula has such an RBM. The construction gives each conjunctive clause its own hidden unit; the clause contributes $-\varepsilon$ to the minimized energy exactly when all its positive literals are true and all its negative literals are false, and $0$ otherwise, so the number of satisfied clauses is proportional to the negative energy. The paper extends the same idea to CNF, to weighted knowledge bases in the style of penalty logic, and to MaxSAT by minimizing free energy.

Load-bearing premise

The load-bearing premise is that searching the RBM's energy landscape reliably reaches the global minima corresponding to satisfying assignments, which is asserted heuristically and which the paper itself reports can fail for formulas with more than 40 variables.

Editorial extensions

If this is right

  • Any propositional knowledge base can be compiled into an RBM with weights and biases read directly off the formula, and the compilation carries a soundness guarantee: satisfying assignments and energy minima coincide.
  • MaxSAT becomes an energy-minimization problem, solvable by off-the-shelf optimizers such as dual annealing on the RBM's free-energy landscape, without training data or a symbolic SAT solver.
  • A verified LBM module can be placed on top of a convolutional or encoder-decoder network to enforce logical constraints during training, as demonstrated on a semantic image interpretation task.
  • Learning from data plus background knowledge in LBM outperforms a purely symbolic system, a purely neural system, and a state-of-the-art neurosymbolic system on five of seven benchmark datasets.
  • For a class of formulas with millions of possible assignments, LBM finds all satisfying assignments after sampling only about 0.37--0.75% of the search space, suggesting reasoning can be efficient despite exponential worst-case growth.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension is that the sampling efficiency reported here transfers to other structured formula classes only when the energy landscape has no deep local minima; random CNFs beyond 40 variables are predicted to defeat plain Gibbs sampling, which the paper already observes.
  • The construction implies an ordering of all truth assignments by energy, not just a satisfiability test; this ranking could be exploited for approximate reasoning, counting models, or sampling near-solutions, directions the paper mentions but does not develop.
  • If the equivalence holds for weighted knowledge bases, the same hidden-unit construction gives a natural way to revise knowledge during learning: start with weights from the logical rules and let contrastive divergence adjust them, which is what the LBM learning experiments do.
  • The free-energy confidence parameter $c$ provides a continuous knob between a smooth optimization landscape and a sharp satisfiability threshold; an adaptive schedule for $c$ during optimization would be a natural extension of the SAT experiments.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The manuscript, framed as a book chapter, proposes Logical Boltzmann Machines (LBM), an energy-based neurosymbolic framework. It defines a translation from SDNF propositional formulas to RBM energy functions (Theorem 1), argues that any WFF can be represented via full DNF, and claims that inference can be performed by energy minimization, Gibbs sampling, or free-energy optimization. It reports experiments on model coverage, ILP benchmarks, SAT solving, MaxSAT, and a semantic image interpretation task, alongside broad discussion of LLMs, accountability, and future neurosymbolic challenges.

Significance. If the central construction is accepted, the paper offers a clean formal bridge between propositional logic and a restricted Boltzmann machine, extending earlier penalty-logic ideas to a tractable restricted architecture and enabling logical constraints to be inserted as verifiable neural modules. The SDNF-to-RBM construction in Theorem 1 is explicit, simple, and mostly correctly proved once the min_h typo is fixed. The paper also provides worked examples and compares LBM with several neurosymbolic systems. However, the broader practical claims are currently undercut by an invalid CNF conversion, an unsupported sampling equivalence, and a confounded experimental comparison in Section 5, so the manuscript needs substantial revision before the practical claims can be accepted.

major comments (4)
  1. [§3.6.1, Eq. (18) and Eq. (5)] The proposed CNF-to-SDNF conversion is not logically valid. For the clause γ = x1 ∨ x2 (so ST = ∅, SK = {1,2}), Eq. (5) yields (¬x2 ∧ x1) ∨ (¬x1 ∧ x2), which is the XOR of x1 and x2; in particular it evaluates to False on x1 = x2 = 1, where the original clause is True. Consequently, the RBM constructed from Eq. (18) does not in general have its global energy minima at the satisfying assignments of the CNF. Since the SAT experiments in §3.6.4 and the MaxSAT construction in §4 explicitly use this translation, those empirical results are built on an unsound representational step. The correct strict DNF for a clause is obtained by an ordering construction (e.g. x1 ∨ x2 ≡ x1 ∨ (¬x1 ∧ x2)), which the paper itself uses in Example 6; Eq. (5) should be replaced accordingly and the experiments redone.
  2. [§3.2, Theorem 1 and Eq. (13)] The statement of Theorem 1 writes sφ(x) = -E(x), but the proof establishes sφ(x) = -1/ε min_h E(x,h); these are not the same object, since the right-hand side requires minimization over hidden units and normalization by ε. Relatedly, Eq. (13) writes sφ(x) = -1/(cε) min_h E(x,h), which cannot equal 1 for a satisfying assignment when min_h E = -ε. The theorem statement, Eq. (13), and the free-energy expressions in Eq. (12) should be corrected, including the missing minus sign in the denominator of Eq. (12) and the consistent use of the temperature τ in Eq. (11).
  3. [§3.3.1, Lemma 2] Lemma 2 asserts that Gibbs sampling on the constructed RBM is equivalent to searching for a satisfying assignment, but the proof merely notes that energy minimization correlates with satisfaction; it gives no convergence or correctness argument for the sampling process. The paper's own §3.6.4 reports that Gibbs sampling often gets stuck in local minima for formulas with more than 40 variables, and proposes to declare formulas likely unsatisfiable when the free energy does not decrease after 1000 steps. That heuristic is not sound, because a local minimum is not evidence of unsatisfiability. The lemma should be weakened to a heuristic claim, and the coverage experiments in §3.5.1 should not be presented as a general equivalence between Gibbs sampling and model search.
  4. [§5, Table 3] The experimental comparison on the semantic image interpretation task is confounded: the text states that the rule (pt1 ∧ pt2) → (ppo ↔ ppt) was used only by LBM and not by the comparison systems DLN, CNLP, and LTN. The higher AUC reported for LBM in Table 3 can therefore be explained by the additional knowledge given to LBM, not by the LBM mechanism itself. A fair comparison requires either giving the same rule to the baselines or ablating the rule from LBM.
minor comments (4)
  1. [§3.5.1] The acceptance criterion is written as "free energy is lower than or equal to − log(1 + exp(cϵ)" and is missing a closing parenthesis; it should also specify that cϵ denotes c·ε with c=5 and ε=0.5.
  2. [§3.6.1, Eq. (5)] The roles of ST and SK are swapped between the conjunctive-clause setting in Theorem 1 and the clause setting in Eq. (3); using different symbols for the two settings would avoid confusion.
  3. [§3.4] The learning analysis in this section relies on an unproved assumption that for a large confidence value c∞ exactly one hidden unit is activated for each satisfying assignment; this should be stated as a heuristic or proved, since it is used to argue that a solution is found.
  4. [Abstract] The abstract contains the typo "neurosynbolic" in the final sentence; it should read "neurosymbolic".

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the SDNF-to-RBM soundness theorem is proven directly in the text, and the self-cited empirical portions do not support the formal derivation.

full rationale

The central derivation is self-contained. Lemma 1 maps each SDNF conjunctive clause to a product energy term, and Theorem 1 constructs, for each clause, a hidden unit with weights +1/-1 on the clause literals and bias -|ST|+epsilon, proving by direct calculation that min_h E(x,h) equals -epsilon exactly on satisfying assignments and 0 otherwise. This is a soundness proof for the construction, not a prediction or a fit: the RBM energy is built from the syntax of the SDNF, and the truth-conditional semantics is verified after construction, not assumed as an input. The free-energy identity (13) and the CNF clause-count relation (19) are formal consequences of the same construction. The chapter does lean on the authors' prior work [52] for empirical figures and for the statement that Section 3 is based on [52], but the formal theorem is not imported from [52] without proof, so this self-citation is not load-bearing. No fitted parameter is relabeled as a prediction; the coverage experiments report search dynamics of a network whose global minima are already characterized by Theorem 1. Two non-circular caveats should be recorded separately: Lemma 2 asserts rather than proves that Gibbs sampling reaches the characterizing minima, and the CNF shortcut in Section 3.6.1/Eq. (18) is unsound for clauses such as x1∨x2 (it produces XOR), but these are correctness/rigor gaps, not circular reductions.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central construction depends on the SDNF conversion, the standard RBM energy form, and several hand-chosen parameters (ε, c, hidden units, BK fractions). No new physical or conceptual entities are introduced; LBM is a system combining existing RBMs and logic. The most fragile dependency is the unproven equivalence between Gibbs sampling and logical model search.

free parameters (4)
  • epsilon = 0.5 in experiments
    Tuning parameter in the energy construction (0<ε<1); controls the energy gap between satisfied and unsatisfied clauses. Chosen by hand, with values like 0.5 used in all experiments.
  • confidence value c = 5 in coverage experiments; 0.1, 0.5, 1, 5, 10 in free-energy plots
    Scaling factor in the softplus free-energy approximation; the theorems use c→∞, but finite c is needed in sampling and optimization. Not fitted, but chosen per experiment.
  • number of added hidden units = 50 for LBM, 100 for baseline RBM
    Section 3.5.2: 'The number of hidden units added to the LBM is chosen arbitrarily at 50'. This is a model capacity choice that can affect the learning comparisons.
  • percentage of data used to build initial LBM = 2.5% (Mutagenesis, KRK) and 10% (UW-CSE, Alzheimer's)
    Section 3.5.2. Choice of background-knowledge strength is arbitrary and affects learning results.
assumptions (5)
  • standard math Any WFF can be converted to SDNF (or CNF) by truth-table expansion
    Used in Theorem 1 and Section 3.4; the conversion can be exponential, which the paper acknowledges.
  • domain assumption The RBM energy function with binary units defines the stated joint distribution and conditional probabilities
    Standard RBM formalism, Eq. (1) and Eq. (11).
  • ad hoc to paper Gibbs sampling on the constructed RBM converges to models of the formula
    Lemma 2 asserts this equivalence without proof; Section 3.6.4 reports local minima failures for larger formulas, so this is an assumption not a theorem.
  • ad hoc to paper The free-energy limit sφ(x) = lim_{c→∞} -1/(cε) F(x) (Eq. 13) holds and is used for acceptance decisions
    The paper states this formally but does not give a full derivation; experimental acceptance thresholds rely on it.
  • ad hoc to paper Learning analysis assumes a large confidence value c∞ and that one hidden unit is activated per satisfying assignment
    Section 3.4 uses this assumption to argue the RBM can learn the SDNF; the assumption is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reasoning in Neurosymbolic AI." pith.science (2026). https://pith.science/paper/JLJFVLXQ

@misc{pith2026250520313,
  author       = {Pith},
  title        = {Pith review of: Reasoning in Neurosymbolic AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JLJFVLXQ}},
  note         = {Machine review of arXiv:2505.20313}
}
read the original abstract

Knowledge representation and reasoning in neural networks have been a long-standing endeavor which has attracted much attention recently. The principled integration of reasoning and learning in neural networks is a main objective of the area of neurosymbolic Artificial Intelligence (AI). In this chapter, a simple energy-based neurosymbolic AI system is described that can represent and reason formally about any propositional logic formula. This creates a powerful combination of learning from data and knowledge and logical reasoning. We start by positioning neurosymbolic AI in the context of the current AI landscape that is unsurprisingly dominated by Large Language Models (LLMs). We identify important challenges of data efficiency, fairness and safety of LLMs that might be addressed by neurosymbolic reasoning systems with formal reasoning capabilities. We then discuss the representation of logic by the specific energy-based system, including illustrative examples and empirical evaluation of the correspondence between logical reasoning and energy minimization using Restricted Boltzmann Machines (RBM). Learning from data and knowledge is also evaluated empirically and compared with a symbolic, neural and a neurosymbolic system. Results reported in this chapter in an accessible way are expected to reignite the research on the use of neural networks as massively-parallel models for logical reasoning and promote the principled integration of reasoning and learning in deep networks. We conclude the chapter with a discussion of the importance of positioning neurosymbolic AI within a broader framework of formal reasoning and accountability in AI, discussing the challenges for neurosynbolic AI to tackle the various known problems of reliability of deep learning.

Figures

Figures reproduced from arXiv: 2505.20313 by the authors.

Figure 1
Figure 1. An initial Sudoku board and two branches generated by placing a [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Free energy term − log(1 + e cx) for different confidence values c. Each free energy term is associated with a conjunctive clause in the SDNF through the weighted sum P i∈A∪B wijxi + θj . Therefore, if a truth-value as￾signment of xB does not satisfy the formula φ, all energy terms will be close to zero. When φ is satisfied, one free energy term will be − log(1+e cϵ), for a choice of 0 < ϵ < 1 from Theorem 1. Thus, … view at source ↗
Figure 3
Figure 3. Linear correlation between satisfiability of a CNF and minimization [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: An RBM equivalent to the XOR formula (x ⊕ y) ↔ z [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: Percentage coverage as a measure of completeness as sampling pro [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: Time taken by LBM to collect all satisfying assignments compared [PITH_FULL_IMAGE:figures/full_fig_p026_6.png]
Figure 7
Figure 7. Figure 7: Energy function and free-energy function with different [PITH_FULL_IMAGE:figures/full_fig_p030_7.png]
Figure 8
Figure 8. Figure 8: The RBM for the Nixon diamond problem has 4 input neurons [PITH_FULL_IMAGE:figures/full_fig_p033_8.png]
Figure 9
Figure 9. Figure 9: Free energy corresponding to an AND gate with different confidence [PITH_FULL_IMAGE:figures/full_fig_p036_9.png]
Figure 10
Figure 10. Figure 10: Free energy corresponding to an OR gate with different confidence [PITH_FULL_IMAGE:figures/full_fig_p036_10.png]
Figure 11
Figure 11. Figure 11: Statistics for clause satisfaction for (a) Loandra (1089 clauses satisfied [PITH_FULL_IMAGE:figures/full_fig_p037_11.png]
Figure 12
Figure 12. Figure 12: Performance of LBM and Loandra on 6 MaxCut experiments (0 to [PITH_FULL_IMAGE:figures/full_fig_p038_12.png]
Figure 13
Figure 13. Figure 13: Learning and reasoning about the PartOf relation in object images by grounding symbolic concepts into Convolutional Neural Networks and adding a logical layer in the form of a LBM module implementing the rule (p t1 ∧p t2 ) → (p po ↔ p pt). Reasoning in the LBM can inf…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. The Knowledge-Reasoning Dissociation: Fundamental Limitations of LLMs in Clinical Natural Language Inference

    cs.AI 2025-08 reject novelty 6.0 of 10

    Across four clinical inference tasks, six LLMs answer paired knowledge probes at 92% accuracy but the main reasoning tasks at 25%, indicating a systematic knowledge-reasoning gap.

Reference graph

Works this paper leans on

58 extracted references · 53 canonical work pages · cited by 1 Pith paper

  1. [52]

    Tran and Artur d’Avila Garcez

    Son N. Tran and Artur d’Avila Garcez. Neurosymbolic reasoning and learn- ing with restricted boltzmann machines. Proceedings of the AAAI Confer- ence on Artificial Intelligence , 37(5):6558–6565, Jun. 2023

    work page 2023

  2. [1]

    Learning to solve circuit-sat: An unsupervised differentiable approach

    Saeed Amizadeh, Sergiy Matusevych, and Markus Weimer. Learning to solve circuit-sat: An unsupervised differentiable approach. In ICLR, 2019

    2019

  3. [2]

    Logic tensor networks

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

    2022

  4. [3]

    Learning optimal chess strategies

    Michael Bain and Stephen Muggleton. Learning optimal chess strategies. In K. Furukawa, D. Michie, and S. Muggleton, editors, Machine intelligence 13: machine intelligence and inductive learning , pages 291–309. Oxford University Press, Inc., New York, NY, USA, 1994

    work page 1994

  5. [4]

    Applications of maxsat in data analysis

    Jeremias Berg, Antti Hyttinen, and Matti Jarvisalo. Applications of maxsat in data analysis. In Daniel Le Berre and Matti Jarvisalo, editors, Proceed- ings of Pragmatics of SAT 2015 and 2018 , volume 59 of EPiC Series in Computing, pages 50–64. EasyChair, 2019

    work page 2015

  6. [5]

    Besold, Artur d’Avila Garcez, Ernesto Jim´ enez-Ruiz, Roberto Confalonieri, Pranava Madhyastha, and Benedikt Wagner, editors

    Tarek R. Besold, Artur d’Avila Garcez, Ernesto Jim´ enez-Ruiz, Roberto Confalonieri, Pranava Madhyastha, and Benedikt Wagner, editors. Neural- Symbolic Learning and Reasoning - 18th International Conference, NeSy 2024, Barcelona, Spain, September 9-12, 2024, Proceedings, Part I, volume 14979 of Lecture Notes in Computer Science . Springer, 2024

    work page 2024

  7. [6]

    Cohen, Fan Yang, and Kathryn Mazaitis

    William W. Cohen, Fan Yang, and Kathryn Mazaitis. Tensorlog: Deep learning meets probabilistic dbs. CoRR, abs/1707.05390, 2017

    arXiv 2017

  8. [7]

    d’Anjou, M

    A. d’Anjou, M. Gra˜ na, F. J. Torrealdea, and M. C. Hernandez. Solving satisfiability via Boltzmann machines. IEEE Trans. Pattern Anal. Mach. Intell., 15(5):514–521, may 1993

    work page 1993

Show all 58 references
  1. [8]

    d’Avila Garcez, K

    A. d’Avila Garcez, K. Broda, and D. Gabbay. Symbolic knowledge ex- traction from trained neural networks: A sound approach. Artif. Intel. , 125(1–2):155–207, 2001

    2001

  2. [9]

    d’Avila Garcez, L

    A. d’Avila Garcez, L. C. Lamb, and D. M. Gabbay. Neural-Symbolic Cog- nitive Reasoning. Springer, 2009

    2009

  3. [10]

    Hypergraph neural networks with logic clauses

    Jo˜ ao Pedro Gandarela de Souza, Gerson Zaverucha, and Artur d’Avila Garcez. Hypergraph neural networks with logic clauses. In International Joint Conference on Neural Networks, IJCNN 2024, Yokohama, Japan, June 30 - July 5, 2024 , pages 1–8. IEEE, 2024. 46

    2024

  4. [11]

    Donadello, L

    I. Donadello, L. Serafini, and A. S. d’Avila Garcez. Logic tensor networks for semantic image interpretation. In IJCAI-17, pages 1596–1602, 2017

    2017

  5. [12]

    Evans and E

    R. Evans and E. Grefenstette. Learning explanatory rules from noisy data. JAIR, 61:1–64, 2018

    2018

  6. [13]

    Fran¸ ca, G

    M. Fran¸ ca, G. Zaverucha, and A. d’Avila Garcez. Fast relational learning using bottom clause propositionalization with artificial neural networks. Mach. Learning, 94(1):81–104, 2014

    2014

  7. [14]

    Artur d’Avila Garcez and Dov M. Gabbay. Fibring neural networks. InPro- ceedings of the 19th National Conference on Artifical Intelligence, AAAI’04, page 342–347. AAAI Press, 2004

    2004

  8. [15]

    Artur d’Avila Garcez and Lu ´ ıs C. Lamb. Neurosymbolic AI: the 3rd wave. Artif. Intell. Rev. , 56(11):12387–12406, March 2023

    2023

  9. [16]

    Ccn+: A neuro-symbolic framework for deep learning with re- quirements

    Eleonora Giunchiglia, Alex Tatomir, Mihaela C˘ at˘ alina Stoian, and Thomas Lukasiewicz. Ccn+: A neuro-symbolic framework for deep learning with re- quirements. International Journal of Approximate Reasoning , 171:109124,

  10. [17]

    Hernandez, F

    C. Hernandez, F. Albizuri, A. DAnjou, M. Gra˜ na, and Francisco Torre- aldea. Efficient solution of max-sat and sat via higher order boltzmann. Revista Investigaci´ on Operacional, 22, 01 2001

    2001

  11. [18]

    M. C. Hernandez, F. X. Albizuri, A. d’Anjou, M. Gra˜ na, and F. J. Torre- aldea. High-order Boltzmann machines for MAX-SAT and SAT. In Stan Gielen and Bert Kappen, editors, ICANN ’93 , pages 479–479, London,

  12. [19]

    Hinton, Simon Osindero, and Yee-Whye Teh

    Geoffrey E. Hinton, Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief nets. Neural Comput., 18(7):1527–1554, July 2006

    2006

  13. [20]

    Decoding intentions

    Andrew Imbrie, Owen Daniels, and Helen Toner. Decoding intentions. https://cset.georgetown.edu/publication/decoding-intentions/, October 2023. Center for Security and Emerging Technology [Online; accessed 20-Jan-2025]

    2023

  14. [21]

    Thinking, fast and slow

    Daniel Kahneman. Thinking, fast and slow . Farrar, Straus and Giroux, New York, 2011

    2011

  15. [22]

    Finding pre-production vehicle configurations using a maxsat framework

    Tiepelt Marcel Kevin and Singh Tilak Raj. Finding pre-production vehicle configurations using a maxsat framework. In 18th International Configura- tion Workshop , pages 117—-122. ´Ecole des Mines d’Albi-Carmaux, 2016

    2016

  16. [23]

    King, Michael J

    Ross D. King, Michael J. E. Sternberg, and Ashwin Srinivasan. Relat- ing chemical activity to structure: An examination of ilp successes. New Generation Computing, 13(3), Dec 1995. 47

    1995

  17. [24]

    Kirkpatrick, C

    S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi. Optimization by simulated annealing. Science, 220(4598):671–680, 1983

    1983

  18. [25]

    Learning max-sat from contextual examples for combinatorial optimisation

    Mohit Kumar, Samuel Kolb, Stefano Teso, and Luc De Raedt. Learning max-sat from contextual examples for combinatorial optimisation. Proceed- ings of the AAAI Conference on Artificial Intelligence , 34(04):4493–4500, Apr. 2020

    2020

  19. [26]

    Learning algorithms for the classification restricted boltzmann machine

    Hugo Larochelle, Michael Mandel, Razvan Pascanu, and Yoshua Bengio. Learning algorithms for the classification restricted boltzmann machine. J. Mach. Learn. Res., 13(1):643–669, March 2012

    2012

  20. [27]

    Can graph neural networks learn to solve the maxsat problem? (student abstract)

    Minghao Liu, Pei Huang, Fuqi Jia, Fan Zhang, Yuchen Sun, Shaowei Cai, Feifei Ma, and Jian Zhang. Can graph neural networks learn to solve the maxsat problem? (student abstract). Proceedings of the AAAI Conference on Artificial Intelligence , 37(13):16264–16265, Sep. 2023

    2023

  21. [28]

    Dmitry Malioutov and Kuldeep S. Meel. Mlic: A maxsat-based framework for learning interpretable classification rules. August 2018

    2018

  22. [29]

    DeepProbLog: Neural probabilistic logic pro- gramming

    Robin Manhaeve, Sebastijan Dumancic, Angelika Kimmig, Thomas De- meester, and Luc De Raedt. DeepProbLog: Neural probabilistic logic pro- gramming. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa- Bianchi, and R. Garnett, editors, Advances in Neural Information Pro...

    2018

  23. [30]

    Chip War: The Fight for the World’s Most Critical Technol- ogy

    Chris Miller. Chip War: The Fight for the World’s Most Critical Technol- ogy. Scribner, New York, 2022

    2022

  24. [31]

    Gsm-symbolic: Understanding the limi- tations of mathematical reasoning in large language models, 2024

    Iman Mirzadeh, Keivan Alizadeh, Hooman Shahrokhi, Oncel Tuzel, Samy Bengio, and Mehrdad Farajtabar. Gsm-symbolic: Understanding the limi- tations of mathematical reasoning in large language models, 2024

    2024

  25. [32]

    Maxsat-based mcs enumeration

    Antonio Morgado, Mark Liffiton, and Joao Marques-Silva. Maxsat-based mcs enumeration. In Armin Biere, Amir Nahir, and Tanja Vos, editors, Hardware and Software: Verification and Testing , pages 86–101, Berlin, Heidelberg, 2013. Springer Berlin Heidelberg

    2013

  26. [33]

    Closing the neural-symbolic cycle: Knowledge extraction, user intervention and distillation from convolutional neural net- works

    Kwun Ho Ngan, James Phelan, Esma Mansouri-Benssassi, Joe Townsend, and Artur d’Avila Garcez. Closing the neural-symbolic cycle: Knowledge extraction, user intervention and distillation from convolutional neural net- works. In Artur d’Avila Garcez, Tarek R. Besold, Marco Gori, ...

    2023

  27. [34]

    A semantic framework for neu- rosymbolic computation

    Simon Odense and Artur d’Avila Garcez. A semantic framework for neu- rosymbolic computation. Artif. Intell. , 340:104273, 2025. 48

    2025

  28. [35]

    de Penning, A

    L. de Penning, A. d’Avila Garcez, L.C. Lamb, and J-J. Meyer. A neural- symbolic cognitive agent for online learning and reasoning. In IJCAI, pages 1653–1658, 2011

    2011

  29. [36]

    Ac- countability in AI: from principles to industry-specific accreditation.CoRR, abs/2110.09232, 2021

    Chris Percy, Simo Dragicevic, Sanjoy Sarkar, and Artur d’Avila Garcez. Ac- countability in AI: from principles to industry-specific accreditation.CoRR, abs/2110.09232, 2021

    2021 arXiv

  30. [37]

    G. Pinkas. Reasoning, nonmonotonicity and learning in connectionist net- works that capture propositional knowledge. Artif. Intell. , 77(2):203–247, 1995

    1995

  31. [38]

    Symmetric neural networks and propositional logic satisfia- bility

    Gadi Pinkas. Symmetric neural networks and propositional logic satisfia- bility. Neural Comput., 3(2):282–291, June 1991

    1991

  32. [39]

    Markov logic networks

    Matthew Richardson and Pedro Domingos. Markov logic networks. Mach. Learn., 62(1-2):107–136, February 2006

    2006

  33. [40]

    Daniel Selsam, Matthew Lamm, Benedikt B¨ unz, Percy Liang, Leonardo de Moura, and David L. Dill. Learning a SAT solver from single-bit super- vision. In International Conference on Learning Representations , 2019

    2019

  34. [41]

    Learning and reasoning with logic tensor networks

    Luciano Serafini and Artur d’Avila Garcez. Learning and reasoning with logic tensor networks. In AI*IA, pages 334–348, 2016

    2016

  35. [42]

    The curse of recursion: Training on generated data makes models forget

    Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Yarin Gal, Nicolas Paper- not, and Ross Anderson. The curse of recursion: Training on generated data makes models forget. ArXiv, abs/2305.17493, 2023

    2023 arXiv

  36. [43]

    Maximum satisfiabil- ity in software analysis: Applications and techniques

    Xujie Si, Xin Zhang, Radu Grigore, and Mayur Naik. Maximum satisfiabil- ity in software analysis: Applications and techniques. In Rupak Majumdar and Viktor Kuncak, editors, Computer Aided Verification - 29th Inter- national Conference, CA V 2017, Heidelberg, Germany, July 24-2...

    2017

  37. [44]

    Smolensky

    P. Smolensky. Constituent structure and explanation in an integrated con- nectionist/symbolic cognitive architecture. In Connectionism: Debates on Psychological Explanation. 1995

    1995

  38. [45]

    Bioinformatics: Problem Solving Paradigms

    Volker Sperschneider. Bioinformatics: Problem Solving Paradigms . Springer Publishing Company, Incorporated, 2008

    2008

  39. [46]

    Srinivasan

    A. Srinivasan. The Aleph manual. http://www.cs.ox.ac.uk/ activities/machlearn/Aleph/aleph.html, 2007. Accessed: 2021-01-23

    2007

  40. [47]

    Srinivasan, S

    A. Srinivasan, S. H. Muggleton, R.D. King, and M.J.E. Sternberg. Mutage- nesis: Ilp experiments in a non-determinate biological domain. In Proceed- ings of the 4th International Workshop on Inductive Logic Programming, volume 237 of GMD-Studien , pages 217–232, 1994. 49

    1994

  41. [48]

    Neural concept binder

    Wolfgang Stammer, Antonia W¨ ust, David Steinmann, and Kristian Kerst- ing. Neural concept binder. CoRR, abs/2406.09949, 2024

    2024 arXiv

  42. [49]

    Towell and J

    G. Towell and J. Shavlik. Knowledge-based artificial neural networks.Artif. Intel., 70:119–165, 1994

    1994

  43. [50]

    Tran and A

    S. Tran and A. Garcez. Deep logic networks: Inserting and extracting knowledge from deep belief networks. IEEE T. Neur. Net. Learning Syst. , (29):246–258, 2018

    2018

  44. [51]

    Son N. Tran. Compositional neural logic programming. In Zhi-Hua Zhou, editor, Proceedings of the Thirtieth International Joint Conference on Ar- tificial Intelligence, IJCAI-21 , pages 3059–3066. International Joint Con- ferences on Artificial Intelligence Organization, 8 2021...

    2021

  45. [53]

    Donti, Bryan Wilder, and J

    Po-Wei Wang, Priya L. Donti, Bryan Wilder, and J. Zico Kolter. SATNet: Bridging deep learning and logical reasoning using a differentiable satisfi- ability solver. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Mach...

    2019

  46. [54]

    Solving maxsat with matrix multiplication, 2023

    David Warde-Farley, Vinod Nair, Yujia Li, Ivan Lobov, Felix Gimeno, and Simon Osindero. Solving maxsat with matrix multiplication, 2023

    2023

  47. [55]

    Generalized simulated annealing algorithm and its application to the thomson model

    Y Xiang, D.Y Sun, W Fan, and X.G Gong. Generalized simulated annealing algorithm and its application to the thomson model. Physics Letters A , 233(3):216–220, 1997

    1997

  48. [56]

    Differentiable learning of logical rules for knowledge base reasoning

    Fan Yang, Zhilin Yang, and William W Cohen. Differentiable learning of logical rules for knowledge base reasoning. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, edi- tors, Advances in Neural Information Processing Systems 30 , p...

  49. [2024]

    Synergies between Machine Learning and Reasoning

  50. [2328]

    Curran Associates, Inc., 2017. 50

    2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.