Neural network for performing beta-token partitioning in a RETE network

Claim 1

A system for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rule being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequenc of conditions and logical connectors of the rules of the target expert system, the functions of each of the nodes capable of being performed on one of the multiple processors of the computer system, the system for building a RETE network comprising: a. a memory configured to store the target expert system and a set of testing data elements for use in determining the satisfaction of the conditions of the rules; b. a beta-token statistics generator coupled to the memory and configured to create a testing RETE-net from the stored target expert system comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once and to generate processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the testing RETE network; c. a programmed neural network coupled to the beta-token statistics generator and configured to generate indicators that identify nodes in the target expert system for beta-token partitioning based on the processing statistics from the beta-token statistics generator. d. an allocator arranged to allocate the nodes and the node memory associated with the join nodes of the testing RETE network to partitions as a function of the indicators generated by the programmed neural network, each partition representing one of the multiple processors; and e. a RETE network builder coupled to the allocator to convert the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.

Claim 2

A system for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rule being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequenc of conditions and logical connectors of the rules of the target expert system, the functions of each of the nodes capable of being performed on one of the multiple processors of the computer system, the system for building a RETE network comprising: a. a memory configured to store the target expert system and a set of testing data elements for use in determining the satisfaction of the conditions of the rules; b. a beta-token statistics generator coupled to the memory and configured to create a testing RETE-net from the stored target expert system comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once and to generate processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the testing RETE network; c. a programmed neural network coupled to the beta-token statistics generator and configured to generate indicators that identify nodes in the target expert system for beta-token partitioning based on the processing statistics from the beta-token statistics generator. d. an allocator arranged to allocate the nodes and the node memory associated with the join nodes of the testing RETE network to partitions as a function of the indicators generated by the programmed neural network, each partition representing one of the multiple processors; and e. a RETE network builder coupled to the allocator to convert the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.

Claim 3

A system for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rule being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions of each of the nodes capable of being performed on one of the multiple processors of the computer system, the system for building a RETE network comprising: a. a memory configured to store the target expert system and store a set of testing data elements for use in determining the satisfaction of the conditions of the rules; b. a beta-token statistics generator coupled to the memory and configured to create a testing RETE-net from the stored target expert system comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once and to generate testing processing statistics for join nodes of the testing RETE network by applying the set of testing data elements to the testing RETE network; c. a neural network programming data provider that provides a neural network with programming data so that the neural network can be programmed to generate beta indicators identifying nodes in the target expert system for beta-token partitioning based on the testing processing statistics; d. a programmed neural network coupled to the beta-token statistics generator and programmed with the data from the neural network programming data provider to generate beta indicators that identify nodes in the target expert system for beta-token partitioning based on the processing statistics from the beta-token statistics generator; e. an allocator arranged to allocate the nodes and the node memory associated with the join nodes of the testing RETE network to partitions as a function of the indicators generated by the programmed neural network, each partition representing one of the multiple processors; and f. a RETE network builder coupled to the allocator to convert the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding processor.

Claim 4

A system for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rule being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions of each of the nodes capable of being performed on one of the multiple processors of the computer system, the system for building a RETE network comprising: a. a memory configured to store the target expert system and store a set of testing data elements for use in determining the satisfaction of the conditions of the rules; b. a beta-token statistics generator coupled to the memory and configured to create a testing RETE-net from the stored target expert system comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once and to generate testing processing statistics for join nodes of the testing RETE network by applying the set of testing data elements to the testing RETE network; c. a neural network programming data provider that provides a neural network with programming data so that the neural network can be programmed to generate beta indicators identifying nodes in the target expert system for beta-token partitioning based on the testing processing statistics; d. a programmed neural network coupled to the beta-token statistics generator and programmed with the data from the neural network programming data provider to generate beta indicators that identify nodes in the target expert system for beta-token partitioning based on the processing statistics from the beta-token statistics generator; e. an allocator arranged to allocate the nodes and the node memory associated with the join nodes of the testing RETE network to partitions as a function of the indicators generated by the programmed neural network, each partition representing one of the multiple processors; and f. a RETE network builder coupled to the allocator to convert the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding processor.

Claim 13

A method for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules, each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rules being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions each of the nodes capable of being performed on one of the multiple processors of the computer system, the method for building a RETE network comprising the steps of: a. storing the target expert system in a memory; b. storing in the memory a set of testing data elements for use in determining the satisfaction of the conditions of the rules; c. utilizing the target expert system in the memory to create a testing RETE network from the stored target expert system, the RETE network comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once, the join nodes having node memory associated therewith; d. generating processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the RETE network; e. generating indicators for beta-token partitioning by applying the target expert system and processing statistics to a programmed neural network; g. allocating the nodes and the node memory associated with the join nodes of the testing RETE network to partitions based on indicators generated by the programmed neural network, each partition representing one of the multiple processors; and h. converting the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.

Claim 14

A method for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules, each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rules being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions each of the nodes capable of being performed on one of the multiple processors of the computer system, the method for building a RETE network comprising the steps of: a. storing the target expert system in a memory; b. storing in the memory a set of testing data elements for use in determining the satisfaction of the conditions of the rules; c. utilizing the target expert system in the memory to create a testing RETE network from the stored target expert system, the RETE network comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once, the join nodes having node memory associated therewith; d. generating processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the RETE network; e. generating indicators for beta-token partitioning by applying the target expert system and processing statistics to a programmed neural network; g. allocating the nodes and the node memory associated with the join nodes of the testing RETE network to partitions based on indicators generated by the programmed neural network, each partition representing one of the multiple processors; and h. converting the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.

Claim 15

A method for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules, each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rules being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions each of the nodes capable of being performed on one on the multiple processors of the computer system, the method for building a RETE network comprising the steps of: a. storing the target expert system in a memory; b. storing in the memory a set of testing data elements for use in determining the satisfaction of the conditions of the rules; c. utilizing the target expert system in the memory to create a testing RETE network from the stored target expert system, the RETE network comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once; d. generating processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the RETE network; e. providing a neural network with programming data so that the neural network can be programmed to generate indicators identifying nodes in the target expert system for beta-token partitioning based on processing statistics; f. generating indicators for beta-token partitioning by applying the target expert system and processing statistics to the programmed neural network; g. allocating the nodes and the node memory associated with the join nodes of the testing RETE network to partitions based on indicators generated by the neural network, each partition representing one of the multiple processors; and h. converting the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.

Claim 16

A method for building a reticular discrimination network (RETE network) capable of being partitioned across a set of multiple, interconnected computer processors comprising a computer system, from at least a portion of a target expert system, the target expert system comprising a plurality of rules, each of which comprises a plurality of conditions grouped in a sequence by a set of logical connectors and a result to occur when the conditions and the logical connectors of the rule are satisfied, the conditions and logical connectors of the rules being satisfied in comparison against a set of data elements, certain ones of the plurality of rules having conditions that are equivalent to conditions in other ones of the plurality of rules, certain ones of the plurality of rules having sequences of conditions and logical connectors that are equivalent to sequences of conditions and logical connectors in other ones of the plurality of the rules, each of the conditions in a rule being represented by a test node and each of the logical connectors being represented by a join node in the RETE network, each join node having node memory associated therewith, the test and join nodes being arranged to determine the satisfaction of the conditions and operation of the logical connectors as represented by the nodes, the nodes also being coupled together to represent the sequence of conditions and logical connectors of the rules of the target expert system, the functions each of the nodes capable of being performed on one on the multiple processors of the computer system, the method for building a RETE network comprising the steps of: a. storing the target expert system in a memory; b. storing in the memory a set of testing data elements for use in determining the satisfaction of the conditions of the rules; c. utilizing the target expert system in the memory to create a testing RETE network from the stored target expert system, the RETE network comprising a set of test nodes and join nodes representing the conditions and logical connectors in the rules of the target expert system arranged such that shared sequences of conditions and logical connectors are represented only once; d. generating processing statistics for each join node of the testing RETE network by applying the set of testing data elements to the RETE network; e. providing a neural network with programming data so that the neural network can be programmed to generate indicators identifying nodes in the target expert system for beta-token partitioning based on processing statistics; f. generating indicators for beta-token partitioning by applying the target expert system and processing statistics to the programmed neural network; g. allocating the nodes and the node memory associated with the join nodes of the testing RETE network to partitions based on indicators generated by the neural network, each partition representing one of the multiple processors; and h. converting the rules of the target expert system into a computer-executable RETE network, according to the partitions, each partition of the RETE network being arranged to execute on a corresponding one of the multiple processors.