## Abstract

Machine learning and logical reasoning have been the two foundational pillars of Artificial Intelligence (AI) since its inception, and yet, until recently the interactions between these two fields have been relatively limited. Despite their individual success and largely independent development, there are new problems on the horizon that seem solvable only via a combination of ideas from these two fields of AI. These problems can be broadly characterized as follows: how can learning be used to make logical reasoning and synthesis/verification engines more efficient and powerful, and in the reverse direction, how can we use reasoning to improve the accuracy, generalizability, and trustworthiness of learning. In this perspective paper, we address the above-mentioned questions with an emphasis on certain paradigmatic trends at the intersection of learning and reasoning. Our intent here is not to be a comprehensive survey of all the ways in which learning and reasoning have been combined in the past. Rather we focus on certain recent paradigms where *corrective feedback loops* between learning and reasoning seem to play a particularly important role. Specifically, we observe the following three trends: first, the use of learning techniques (especially, reinforcement learning) in sequencing, selecting, and initializing proof rules in solvers/provers; second, combinations of inductive learning and deductive reasoning in the context of program synthesis and verification; and third, the use of solver layers in providing corrective feedback to machine learning models in order to help improve their accuracy, generalizability, and robustness with respect to partial specifications or domain knowledge. We believe that these paradigms are likely to have significant and dramatic impact on AI and its applications for a long time to come.

## Notes

Given that the focus of this paper is to provide a perspective on certain emerging trends in AI, there is no experimental or scientific data associated with it.

Conceptually, the terms abstraction and approximation are very similar in their technical meaning in the broad context of program analysis, synthesis, and verification. Hence, we use them interchangeably.

It is possible to imagine more complex configurations with multiple kinds of abstractors and refiners, but we won’t discuss them for the sake of brevity.

In this paper, we interchangeably use the terms logical reasoners, symbolic reasoners, decision procedures, and solvers (dually, provers), since all such systems fundamentally reason about mathematical formulas and decide whether they are satisfiable (dually, valid).

## References

Alur R, Bodik R, Juniwal G, Martin Milo MK, Raghothaman M, Seshia SA, Singh R, Solar-Lezama A, Torlak E, Udupa A (2013) Syntax-guided synthesis. In: Proceedings of the IEEE international conference on formal methods in computer-aided design (FMCAD), pp 1–17

Abate A, David C, Kesseli P, Kroening D, Polgreen E (2018) Counterexample guided inductive synthesis modulo theories. In: Computer aided verification-30th international conference (CAV), volume 10981 of Lecture Notes in Computer Science, pp 270–288. Springer

Atserias A, Fichte JK, Thurley M (2009) Clause-learning algorithms with many restarts and bounded-width resolution. In: Kullmann O (ed) Theory and Applications of Satisfiability Testing - SAT. Springer, Berlin Heidelberg, Berlin, Heidelberg, pp 114–127

Angluin D (1988) Queries and concept learning. Mach Learn 2(4):319–342

Audemard Gilles, Simon Laurent (2013) Glucose 2.3 in the SAT 2013 Competition. In: Proceedings of SAT competition 2013, pp 42–43

Alur R, Singh R, Fisman D, Solar-Lezama A (2018) Search-based program synthesis. Commun ACM 61(12):84–93

Ashok D, Scott J, Wetzel SJ, Panju M, Ganesh V (2021) Logic guided genetic algorithms. In: 35th AAAI conference on artificial intelligence, AAAI 2021, 33rd Conference on innovative applications of artificial intelligence, IAAI 2021, The 11th symposium on educational advances in artificial intelligence, EAAI 2021, virtual event, Feb 2–9, 2021, pp 15753–15754. AAAI Press

Alaa AM, Schaar M van der (2019) Demystifying black-box models with symbolic metamodels. Adv Neural Inf Process Syst 32

Bouraoui Z, Cornuéjols A, Denoeux T, Destercke S, Dubois D, Guillaume R, Marques-Silva J, Mengin J, Prade H, Schockaert S, Serrurier M, Vrain C (2019) From shallow to deep interactions between knowledge representation, reasoning and machine learning (kay r. amel group). CoRR, abs/1912.06612

Biere A, Heule M, van Maaren H, Walsh T (2009) (eds) Handbook of Satisfiability, vol. 185 of Frontiers in Artificial Intelligence and Applications. IOS Press

Bünz B, Lamm M (2017) Graph neural networks and boolean satisfiability. CoRR, abs/1702.03592

Bengio Y, LeCun Y, Hinton GE (2021) Deep learning for AI. Commun ACM 64(7):58–65

Bak S, Liu C, Johnson TT (2021) The second international verification of neural networks competition (VNN-COMP 2021): Summary and results. CoRR, abs/2109.00498

Ball T, Levin V, Rajamani SK (2011) A decade of software model checking with SLAM. Commun ACM 54(7):68–76

Bansal K, Loos SM, Rabe MN, Szegedy C, Wilcox S (2019) Holist: an environment for machine learning of higher order logic theorem proving. In: Kamalika C and Ruslan S, (eds.) Proceedings of the 36th International Conference on Machine Learning, ICML 2019, 9-15 June 2019, Long Beach, California, USA, volume 97 of Proceedings of Machine Learning Research, pp 454–463. PMLR

Bader-El-Den MB, Poli R (2007) Generating SAT local-search heuristics using a GP hyper-heuristic framework. In: Nicolas M, El-Ghazali T, Pierre C, Marc S, Evelyne L, (eds), Artificial Evolution, 8th International Conference, Evolution Artificielle, EA 2007, Tours, France, October 29-31, 2007, revised selected papers, volume 4926 of Lecture Notes in Computer Science, Springer, pp 37–49

Clarke E, Biere A, Raimi R, Zhu Y (2001) Bounded model checking using satisfiability solving. Formal Methods Syst Des 19(1):7–34

Cropper A, Dumancic S, Evans R, Muggleton SH (2022) Inductive logic programming at 30. Mach Learn 111(1):147–172

Clarke EM, Emerson EA (1981) Design and synthesis of synchronization skeletons using branching-time temporal logic. In: Logic of programs, pp 52–71

Clarke EM, Fehnker A, Han Z, Krogh BH, Stursberg O, Theobald M (2003) Verification of hybrid systems based on counterexample-guided abstraction refinement. In: TACAS, pp 192–207

Clarke EM, Grumberg O, Jha S, Lu Y, Veith H (2000) Counterexample-guided abstraction refinement. In: 12th international conference on computer aided verification (CAV), vol. 1855 of Lecture Notes in Computer Science, Springer, pp 154–169

Clarke EM, Grumberg O, Jha S, Yuan L, Veith H (2003) Counterexample-guided abstraction refinement for symbolic model checking. J ACM 50(5):752–794

Clarke EM, Gupta A, Kukula JH, Strichman O (2002) SAT based abstraction-refinement using ILP and machine learning techniques. In: computer aided verification, 14th international conference (CAV), vol. 2404 of Lecture Notes in Computer Science, Springer, pp 265–279

Cadar C, Ganesh V, Pawlowski PM, Dill DL, Engler DR (2006) EXE: automatically generating inputs of death. In: Proceedings of the 13th ACM conference on computer and communications security, CCS ’06, New York, NY, USA. ACM, pp 322–335

Cook S (1971) The complexity of theorem-proving procedures. In: proceedings of the third annual ACM symposium on theory of computing (STOC), ACM, pp 151–158

Dash T, Chitlangia S, Ahuja A, Srinivasan A (2021) How to tell deep neural networks what we know. CoRR, abs/2107.10295

Davis M, Logemann G, Loveland D (1962) A machine program for theorem-proving. Commun ACM 5(7):394–397

De Moura L, Bjørner N (2008) Z3: An efficient smt solver. In: international conference on tools and algorithms for the construction and analysis of systems, Springer, pp 337–340

Duan H, Nejati S, Trimponias G, Poupart P, Ganesh V (2020) Online bayesian moment matching based SAT solver heuristics. In: Proceedings of the 37th international conference on machine learning, ICML 2020, 13–18 July 2020, Virtual Event, vol. 119 of proceedings of machine learning research, PMLR, pp 2710–2719

Devlin D, O’Sullivan B (2008) Satisfiability as a classification problem. In: Proceedings of the 19th Irish Conference on artificial intelligence and cognitive science

Dubois D, Prade H (2019) Towards a reconciliation between reasoning and learning-a position paper. In: Nahla Ben A, Benjamin Q, Martin T, (eds) scalable uncertainty management-13th international conference, SUM 2019, Compiègne, France, Dec 16–18, 2019, Proceedings, vol. 11940 of lecture notes in computer science, Springer, pp 153–168

Eén N, Sörensson N (2004) Theory and applications of satisfiability testing: 6th international conference, SAT 2003, santa margherita ligure, Italy, May 5–8, 2003, selected revised papers, chapter an extensible SAT-solver, Springer Berlin Heidelberg, Berlin, Heidelberg, pp 502–518

Ertel W, Schumann J, Suttner CB (1989) Learning heuristics for a theorem prover using back propagation. In: Johannes R, Karl L, (eds) 5. Österreichische Artificial Intelligence-Tagung, Igls, Tirol, 28. bis 30. Sept 1989, Proceedings, vol. 208 of Informatik-Fachberichte, Springer, pp 87–95

Flint A, Blaschko MB (2012) Perceptron learning of SAT. In: Bartlett PL, Pereira FNC, Burges CJC, Léon B, Weinberger KQ, (eds) Advances in neural information processing systems 25: 26th annual conference on neural information processing systems 2012. Proceedings of a meeting held Dec 3–6, 2012, Lake Tahoe, Nevada, United States, pp 2780–2788

Froleyks N, Heule M, Iser M, Järvisalo M, Suda M (2021) SAT competition 2020. Artif Intell 301:103572

Feldman YMY, Immerman N, Sagiv M, Shoham S (2020) Complexity and information in invariant inference. Proc ACM Program Lang 4(POPL):5:1-5:29

First E, Rabe MN, Ringer T, Brun Y (2023) Baldur: whole-proof generation and repair with large language models. CoRR, abs/2303.04910

Fukunaga AS (2002) Automated discovery of composite SAT variable-selection heuristics. In: Dechter R, Kearns MJ, Sutton RS, (eds) proceedings of the eighteenth national conference on artificial intelligence and fourteenth conference on innovative applications of artificial intelligence, July 28 - August 1, 2002, Edmonton, Alberta, Canada, AAAI Press / The MIT Press, pp 641–648

Fukunaga AS (2004) Evolving local search heuristics for SAT using genetic programming. In: Kalyanmoy D, Riccardo P, Wolfgang B, Hans-Georg B, Burke EK, Darwen PJ, Dasgupta D, Floreano D, Foster JA, Mark H, Owen H, Pier Luca L, Lee S, Andrea T, Dirk T, Tyrrell AM, (eds) Genetic and evolutionary computation-GECCO 2004, genetic and evolutionary computation conference, Seattle, WA, USA, June 26–30, 2004, Proceedings, Part II, volume 3103 of Lecture Notes in Computer Science, Springer, pp 483–494

Fukunaga AS (2008) Automated discovery of local search heuristics for satisfiability testing. Evol Comput 16(1):31–61

Ian JG, Yoshua B, Courville AC (2016) Deep Learning. Adaptive computation and machine learning. MIT Press, Cambridge

Goldman SA, Kearns MJ (1992) On the complexity of teaching. J Comput Syst Sci 50:303–314

Gupta A (2006) Learning Abstractions for Model Checking. PhD thesis, Carnegie Mellon University, June

Hutter F, Babic D, Hoos HH, Hu AJ (2007) Boosting verification by automatic tuning of decision procedures. In: 7th international conference on formal methods in computer-aided design (FMCAD), IEEE Computer Society, pp 27–34

Hart PE, Nilsson NJ, Raphael B (1968) A formal basis for the heuristic determination of minimum cost paths. IEEE Trans Syst Sci Cybern 4(2):100–107

Holden SB (2021) Machine learning for automated theorem proving: learning to solve SAT and QSAT. Found Trends Mach Learn 14(6):807–989

Hopfield JJ, Tank DW (1985) Neural computation of decisions in optimization problems. Biol Cybern 52:141–152

Jha S, Gulwani S, Seshia SA, Tiwari A (2010) Oracle-guided component-based program synthesis. In: Proceedings of the 32nd international conference on software engineering (ICSE), pp 215–224

Jha S, Gulwani S, Seshia SA, Tiwari A (2010) Synthesizing switching logic for safety and dwell-time requirements. In: proceedings of the international conference on cyber-physical systems (ICCPS), pp 22–31

Johnson JL (1989) A neural network approach to the 3-satisfiability problem. J Parallel Distributed Comput 6:435–449

Jha S, Seshia SA (2017) A theory of formal synthesis via inductive learning. Acta Inform 54(7):693–726

Kim S, Lu PY, Mukherjee S, Gilbert M, Jing L, Čeperić V, Soljačić M (2021) Integration of neural network-based symbolic regression in deep learning for scientific discovery. IEEE Trans Neural Netw Learn Syst 32(9):4166–4177

Kautz HA, Selman B (1992) Planning as satisfiability. In: Bernd N, (ed) 10th European conference on artificial intelligence, ECAI 92, Vienna, Austria, August 3–7, 1992. Proceedings, John Wiley and Sons, pp 359–363

Kurshan R (1994) Automata-theoretic verification of coordinating processes. In: 11th international conference on analysis and optimization of systems–discrete event systems, vol. 199 of LNCS, Springer, pp 16–28

Lewkowycz A, Andreassen A, Dohan D, Dyer E, Michalewski H, Ramasesh VV, Slone A, Anil C, Schlag I, Gutman-Solo T, Wu Y, Neyshabur B, Gur-Ari G, Misra V (2022) Solving quantitative reasoning problems with language models. CoRR, abs/2206.14858

Liang JH, Ganesh V, Poupart P, Czarnecki K (2016) Learning rate based branching heuristic for SAT solvers. In: Nadia C, Daniel LB, (eds) Theory and applications of satisfiability testing–SAT 2016, Cham. Springer International Publishing, pp 123–140

Liang JH (2018) Machine learning for SAT solvers. PhD thesis, University of Waterloo, Canada

Lagoudakis Michail G, Littman Michael L (2001) Learning to select branching rules in the DPLL procedure for satisfiability. Electron Notes Discrete Math 9:344–359

Liang JH, Oh C, Mathew M, Thomas C, Li C, Ganesh V (2018) Machine learning-based restart policy for CDCL SAT solvers. In: theory and applications of satisfiability testing-SAT 2018 - 21st international conference, SAT 2018, held as part of the federated logic conference, FloC 2018, Oxford, UK, July 9–12, 2018, Proceedings, pp 94–110

Lederman G, Rabe MN, Seshia S, Lee EA (2020) Learning heuristics for quantified boolean formulas through reinforcement learning. In: 8th international conference on learning representations (ICLR), April

Lorenz JH, Wörz F (2020) On the effect of learned clauses on stochastic local search. In: Luca P, Martina S, (eds) Theory and applications of satisfiability testing-SAT 2020–23rd international conference, Alghero, Italy, July 3–10, 2020, Proceedings, vol. 12178 of lecture notes in computer science, Springer, pp 89–106

Tom M (1997) Mitchell. McGraw-Hill, Machine Learning

Marques-Silva JP, Sakallah KA (1996) GRASP-A new search algorithm for satisfiability. In: proceedings of the 1996 IEEE/ACM international conference on computer-aided design, ICCAD ’96, Washington, DC, USA. IEEE Computer Society, pp 220–227

Manna Z, Waldinger R (1980) A deductive approach to program synthesis. ACM Trans Program Lang Syst (TOPLAS) 2(1):90–121

Nejati S, Frioux LL, Ganesh V (2020) A machine learning based splitting heuristic for divide-and-conquer solvers. In: Helmut S, (ed) Principles and practice of constraint programming-26th international conference, CP 2020, Louvain-la-Neuve, Belgium, Sept 7–11, 2020, Proceedings, vol. 12333 of lecture notes in computer science, Springer, pp 899–916

OpenAI. GPT-4 technical report. CoRR, abs/2303.08774, 2023

Panju M (2021) Automated Knowledge Discovery using Neural Networks. PhD thesis, University of Waterloo, Ontario, Canada

Polgreen E, Cheang K, Gaddamadugu P, Godbole A, Laeufer K, Lin S, Manerkar YA, Mora F, Seshia SA (2022) UCLID5: multi-modal formal modeling, verification, and synthesis. In: computer aided verification-34th international conference (CAV), vol. 13371 of lecture notes in computer science, Springer, pp 538–551

Pipatsrisawat K, Darwiche A (2011) On the power of clause-learning SAT solvers as resolution engines. Artif Intell 175(2):512–525

Andrei P, Polat ES, Alexander F, Mathias U, Müslüm A, Viet-Man L, Klaus P, Martin E, Trang TTN (2022) An overview of machine learning techniques in constraint solving. J Intell Inf Syst 58(1):91–118

Pimpalkhare N, Mora F, Polgreen E, Seshia SA (2021) MedleySolver: online SMT algorithm selection. In: 24th international conference on theory and applications of satisfiability testing (SAT), vol. 12831 of lecture notes in computer science, Springer, pp 453–470

Pogancic MV, Paulus A, Musil V, Martius G, Rolínek M (2020) Differentiation of blackbox combinatorial solvers. In: 8th international conference on learning representations, ICLR 2020, Addis Ababa, Ethiopia, April 26–30, 2020. OpenReview.net

Polgreen E, Reynolds A, Seshia SA (2022) Satisfiability and synthesis modulo oracles. In: Proceedings of the 23rd international conference on verification, model checking, and abstract interpretation (VMCAI)

De Raedt L (2008) Logical and relational learning. Cognitive technologies. Springer, Berlin

Russell S, Norvig P (2020) Artificial intelligence: a modern approach, 4th edn. Pearson, London

Rossi F, van Beek P, Walsh T (2006) editors. Handbook of Constraint Programming, vol. 2 of Foundations of Artificial Intelligence. Elsevier

Richard SS, Andrew GB (1998) Reinforcement learning-an introduction. Adaptive computation and machine learning. MIT Press, Cambridge

Seshia SA (2005) Adaptive eager boolean encoding for arithmetic reasoning in verification. PhD thesis, Carnegie Mellon University

Seshia SA (2012) Sciduction: combining induction, deduction, and structure for verification and synthesis. In: proceedings of the design automation conference (DAC), pp 356–365

Seshia SA (2015) Combining induction, deduction, and structure for verification and synthesis. Proc IEEE 103(11):2036–2051

Selsam D, Lamm M, Bünz B, Liang P, de Moura L, Dill DL (2019) Learning a SAT solver from single-bit supervision. In: 7th international conference on learning representations, ICLR 2019, New Orleans, LA, USA, May 6–9, 2019. OpenReview.net

Solar-Lezama A, Tancau L, Bodík R, Seshia SA, Saraswat VA (2006) Combinatorial sketching for finite programs. In: proceedings of the 12th international conference on architectural support for programming languages and operating systems (ASPLOS), ACM Press, pp 404–415

Scott J, Niemetz A, Preiner M, Nejati S, Ganesh V (2021) Machsmt: a machine learning-based algorithm selector for SMT solvers. In: Jan Friso G, Kim Guldstrand L, (eds) Tools and algorithms for the construction and analysis of systems-27th international conference, TACAS 2021, Held as Part of the European Joint Conferences on Theory and Practice of Software, ETAPS 2021, Luxembourg City, Luxembourg, March 27–April 1, 2021, Proceedings, Part II, vol. 12652 of Lecture Notes in Computer Science, Springer, pp 303–325

Scott J, Panju M, Ganesh V (2020) LGML: logic guided machine learning. In: the thirty-fourth AAAI conference on artificial intelligence, AAAI 2020, the thirty-second innovative applications of artificial intelligence conference, IAAI 2020, the tenth AAAI symposium on educational advances in artificial intelligence, EAAI 2020, New York, NY, USA, Feb 7–12, 2020, AAAI Press, pp 13909–13910

Marques SJP, Sakallah KA (1996) GRASP-a new search algorithm for satisfiability. In: Rutenbar RA, Otten RHJM, (eds) proceedings of the 1996 IEEE/ACM international conference on computer-aided design, ICCAD 1996, San Jose, CA, USA, Nov 10–14, 1996, IEEE Computer Society / ACM, pp 220–227

Seshia SA, Subramanyan P (2018) UCLID5: integrating modeling, verification, synthesis, and learning. In: proceedings of the 15th ACM/IEEE international conference on formal methods and models for codesign (MEMOCODE)

Seshia SA, Sadigh D, Sastry SS (2022) Toward verified artificial intelligence. Commun ACM 65(7):46–55

Seshia SA, Sharygina N, Tripakis S (2018) Modeling for verification. In: Clarke EM, Thomas H, Helmut V, (eds) Handbook of Model Checking, chapter 3. Springer

Sarker Md, Kamruzzaman ZL, Aaron E, Pascal H (2021) Neuro-symbolic artificial intelligence. AI Commun 34(3):197–209

van Harmelen F, Lifschitz V, Porter BW (2008) (eds) Handbook of Knowledge Representation, vol. 3 of Foundations of Artificial Intelligence. Elsevier

The Verification of Neural Networks Library (VNN-LIB). www.vnnlib.org, 2019

Wang PW, Donti PL, Wilder B, Zico KJ (2019) Satnet: bridging deep learning and logical reasoning using a differentiable satisfiability solver. In: Kamalika C, Ruslan S, (eds) proceedings of the 36th international conference on machine learning, ICML 2019, 9–15 June 2019, Long Beach, California, USA, vol. 97 of Proceedings of Machine Learning Research, PMLR, pp 6545–6554

Jeannette MW (2021) Trustworthy AI. Commun ACM 64(10):64–71

Lin X, Hutter F, Hoos HH, Leyton-Brown K (2008) SATzilla: portfolio-based algorithm selection for SAT. J Artif Intell Res 32(1):565–606

## Author information

### Authors and Affiliations

### Corresponding author

## Additional information

### Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

## About this article

### Cite this article

Ganesh, V., Seshia, S.A. & Jha, S. Machine learning and logic: a new frontier in artificial intelligence.
*Form Methods Syst Des* **60**, 426–451 (2022). https://doi.org/10.1007/s10703-023-00430-1

Received:

Accepted:

Published:

Issue Date:

DOI: https://doi.org/10.1007/s10703-023-00430-1