Granted Patent B2

US 12,608,589 · App. 17/807,909 · Granted Apr 21, 2026

Inventors: Ndivhuwo Makondo (Pretoria, ZA); Francois Pierre Luus (Wierdapark, ZA); Naweed Aghmad Khan (Johannesburg, ZA); Ismail Yunus Akhalwaya (Emmarentia, ZA); Ryan Nelson Riegel (Carrollton, GA); Oarabile Hope Moloko (Mahikeng, ZA); Thabang Doreen Lebese (Johannesburg, ZA)

Assignee: International Business Machines Corporation

G06N3/042G06N5/046

Abstract

Claims (79)

1 . A computer-implemented method comprising:

receiving, by one or more processors, input data corresponding to a knowledge base, the input data including a plurality of propositional logic clauses;

generating, by one or more processors, output data corresponding to a first set of logical rules and a first set of facts based on the plurality of propositional logic clauses;

accumulating, by one or more processors, the first set of facts to generate accumulated facts;

generating, by one or more processors, a logical neural network (LNN) graph based on the accumulated facts and the first set of logical rules, wherein the LNN graph comprises neurons corresponding to the plurality of propositional logic clauses and provides one-to-one correspondence between the neurons and logical elements represented in the plurality of propositional logic clauses;

alternating, by one or more processors, reasoning and learning passes at the LNN graph until convergence to generate a second set of logical rules and a second set of facts;

generating, by one or more processors, a third set of logical rules and a third set of facts, wherein the third set of logical rules and the third set of facts exceed a first predetermined threshold; and

correcting, by one or more processors, errors in the knowledge base by updating the knowledge base to include the third set of logical rules and the third set of facts.

2 . The computer-implemented method of claim 1 , wherein the input data corresponding to the knowledge base further comprises an external labeled dataset aligned with the knowledge base.

3 . The computer-implemented method of claim 1 , wherein generating the output data further comprises:

receiving, by one or more processors, the input data;

processing, by one or more processors, the input data to identify each of the plurality of propositional logic clauses;

extracting, by one or more processors, the first set of logical rules and the first set of facts from the plurality of propositional logic clauses; and

transmitting, by one or more processors, the first set of logical rules and the first set of facts as the output data.

4 . The computer-implemented method of claim 1 , wherein generating the LNN graph further comprises:

generating, by one or more processors, a graphical representation of first order logic statements corresponding to syntax trees representing the accumulated facts with the corresponding first set of logical rules; and

displaying, by one or more processors, the graphical representation as the LNN graph on a user interface of a computing device.

5 . The computer-implemented method of claim 1 , further comprising:

extracting, by one or more processors, truth values from the knowledge base; and

determining, by one or more processors, truth value bounds based on the truth values.

6 . The computer-implemented method of claim 5 , wherein alternating the reasoning and learning passes at the LNN graph until convergence further comprises:

applying, by one or more processors, the truth value bounds to the LNN graph to allow an open world assumption that some of the second set of logical rules and the second set of facts are true even if the second set of logical rules and the second set of facts are unknown and unprovable.

7 . The computer-implemented method of claim 1 , further comprising:

generating, by one or more processors, a fourth set of logical rules and a fourth set of facts, wherein the fourth set of logical rules and the fourth set of facts exceed a second predetermined threshold.

8 . The computer-implemented method of claim 1 , wherein alternating the reasoning and the learning passes at the LNN graph until convergence comprises:

receiving, by one or more processors, training data; and

processing, by one or more processors, the training data by:

performing, by one or more processors, the reasoning by iteratively tightening bounds until convergence, wherein each inference computation, for any node in the LNN graph, results in a selection from the group consisting of: a tighter bounds and remain same; and

subsequent to performing the reasoning, performing, by one or more processors, gradient descent to inform how parameters of the LNN graph are to be updated.

9 . A computer program product comprising:

one or more computer readable storage media; and

program instructions stored on the one or more computer readable storage media to perform operations comprising:

receiving input data corresponding to a knowledge base, the input data including a plurality of propositional logic clauses;

generating output data corresponding to a first set of logical rules and a first set of facts based on the plurality of propositional logic clauses;

accumulating the first set of facts to generate accumulated facts;

generating a logical neural network (LNN) graph based on the accumulated facts and the first set of logical rules, wherein the LNN graph comprises neurons corresponding to the plurality of propositional logic clauses and provides one-to-one correspondence between the neurons and logical elements represented in the plurality of propositional logic clauses;

alternating reasoning and learning passes at the LNN graph until convergence to generate a second set of logical rules and a second set of facts;

generating a third set of logical rules and a third set of facts, wherein the third set of logical rules and the third set of facts exceed a first predetermined threshold; and

correcting errors in the knowledge base by updating the knowledge base to include the third set of logical rules and the third set of facts.

10 . The computer program product of claim 9 , wherein the input data corresponding to the knowledge base further comprises an external labeled dataset aligned with the knowledge base.

11 . The computer program product of claim 9 , wherein the generating the output data further comprises:

receiving the input data;

processing the input data to identify each of the plurality of propositional logic clauses;

extracting the first set of logical rules and the first set of facts from the plurality of propositional logic clauses; and

transmitting the first set of logical rules and the first set of facts as the output data.

12 . The computer program product of claim 9 , wherein the generating the LNN graph further comprises:

generating a graphical representation of first order logic statements corresponding to syntax trees representing the accumulated facts with the corresponding first set of logical rules; and

displaying the graphical representation as the LNN graph on a user interface of a computing device.

13 . The computer program product of claim 9 , wherein the operations further comprise:

extracting truth values from the knowledge base; and

determining truth value bounds based on the truth values.

14 . The computer program product of claim 13 , wherein alternating the reasoning and learning passes at the LNN graph until convergence further comprises:

applying the truth value bounds to the LNN graph to allow an open world assumption that some of the second set of logical rules and the second set of facts are true even if the second set of logical rules and the second set of facts are unknown and unprovable.

15 . The computer program product of claim 9 , wherein the operations further comprise:

generating a fourth set of logical rules and a fourth set of facts, wherein the fourth set of logical rules and the fourth set of facts exceed a second predetermined threshold.

16 . A computer system comprising:

a processor set;

one or more computer readable storage media; and

program instructions stored on the one or more computer readable storage media to cause the processor set to perform operations comprising:

receiving input data corresponding to a knowledge base, the input data including a plurality of propositional logic clauses;

generating output data corresponding to a first set of logical rules and a first set of facts based on the plurality of propositional logic clauses;

accumulating the first set of facts to generate accumulated facts;

generating a logical neural network (LNN) graph based on the accumulated facts and the first set of logical rules, wherein the LNN graph comprises neurons corresponding to the plurality of propositional logic clauses and provides one-to-one correspondence between the neurons and logical elements represented in the plurality of propositional logic clauses;

alternating reasoning and learning passes at the LNN graph until convergence to generate a second set of logical rules and a second set of facts;

generating a third set of logical rules and a third set of facts, wherein the third set of logical rules and the third set of facts exceed a first predetermined threshold; and

correcting errors in the knowledge base by updating the knowledge base to include the third set of logical rules and the third set of facts.

17 . The computer system of claim 16 , wherein the input data corresponding to the knowledge base further comprises an external labeled dataset aligned with the knowledge base.

18 . The computer system of claim 16 , wherein the generating the output data further comprises:

receiving the input data;

processing the input data to identify each of the plurality of propositional logic clauses;

extracting the first set of logical rules and the first set of facts from the plurality of propositional logic clauses; and

transmitting the first set of logical rules and the first set of facts as the output data.

19 . The computer system of claim 16 , wherein the generating the LNN graph further comprises:

generating a graphical representation of first order logic statements corresponding to syntax trees representing the accumulated facts with the corresponding first set of logical rules; and

displaying the graphical representation as the LNN graph on a user interface of a computing device.

20 . The computer system of claim 15 , wherein the operations further comprise:

extracting truth values from the knowledge base; and

determining truth value bounds based on the truth values, wherein alternating the reasoning and learning passes at the LNN graph until convergence further comprises:

applying the truth value bounds to the LNN graph to allow an open world assumption that some of the second set of logical rules and the second set of facts are true even if the second set of logical rules and the second set of facts are not known or provable.

Assignments (1)

Continuity (1)

References Cited (22)