Method of controlling a classifier system

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value for indicating the past performance of said classifier during the course of operation of said adaptive computing system, and
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including:
• means for generating a bid-value whenever the condition part of a given classifier is satisfied, said bid value being related to the strength value of said given classifier,
• means for placing those output messages generated by successful classifiers which produce relatively larger bid-values in said storage means for further processing, and
• means for increasing the strength value associated with said successful classifiers,
• the improvement comprising processing means for varying said bid value in response to the combined strength values of those classifiers which generated said selected messages.

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value for indicating the past performance of said classifier during the course of operation of said adaptive computing system, and
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including:
• means for placing those output messages generated by successful classifiers having larger strength values in said storage means for further processing, and
• means for increasing the strength value associated with said successful classifiers,
• the improvement comprising processing means for periodically reducing the strength value associated with all of said classifiers in said storage means whereby the strength of those classifiers which fail to produce messages with acceptable strength values is eventually diminished relative to others of said classifiers.

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value for indicating the past performance of said classifier during the course of operation of said adaptive computing system, and
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including:
• means for placing those output messages generated by successful classifiers having larger strength values in said storage means for further processing, and
• means for increasing the strength value associated with said successful classifiers;
• and
• an input interface for generating input messages for use in said system derived from a source external to said system;
• the improvement wherein said input interface further includes means for generating an intensity value associated with each of said input messages which is functionally equivalent to a bid value proportional to said strength value associated with classifier-produced messages whereby messages produced by said input means are treated as if they had been generated by a classifier making a comparable bid.

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value indicative of the past performance of said classifier in said system;
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including means for placing selected ones of the output messages generated by classifiers whose condition part is satisfied into said storage means for further processing, and
• means for periodically producing new classifiers which are modified versions of one or more existing classifiers in said storage means:
• the improvement wherein said classifiers include positional punctuation marks imbedded therein and wherein said means for producing new classifiers includes means for combining parts of existing classifiers formed by dividing said existing classifiers at positions determined by the placement of said punctuation marks.

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value indicative of the past performance of said classifier in said system;
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including means for placing selected ones of the output messages generated by classifiers whose condition part is satisfied into said storage means for further processing, and
• means for periodically producing new classifiers having an information content derived in part from selected classifiers in said storage means;
• the improvement wherein said means for producing new classifiers further includes means for deriving the content of new classifiers from the information content of selected messages in said storage means.

In an adaptive computing system of the class comprising, in combination,

• storage means for storing a plurality of messages and a plurality of classifiers, each of said classifiers comprising:
• a condition part which specifies the attributes of selected ones of said messages,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value indicative of the past performance of said classifier in said system;
• processing means for comparing the condition part of each of said classifiers in said storage means with each of said messages in said storage means to identify said selected messages, said processing means further including means for placing selected ones of the output messages generated by classifiers whose condition part is satisfied into said storage means for further processing, and
• means for periodically producing new classifiers which are modified versions of one or more existing classifiers in said message store;
• the improvement wherein said means for producing new classifiers includes means for forming first and second classifiers which operate in succession, the first of said classifiers generating an output message which thereafter satisfies the condition part of the second of said classifiers whereby a message at one time which satisfies the condition part of said first classifier necessarily produces an output message from said second classifier at a later time.

An adaptive computing system comprising, in combination,

• a central computing unit including means for storing a plurality of messages,
• a plurality of independent processing units, each of which includes means for storing a classifier, said classifier comprising
• a condition part which specifies the attributes of selected ones of the messages stored in central computing unit,
• an action part which specifies the functional relationship between the information content of said selected messages and at least one output message, and
• a strength value indicative of the past performance of said classifier;
• parallel switching means for interconnecting said independent processing units;
• broadcast switching means for interconnecting said central computer with each of said independent processing units;
• an input interface for providing data to said central computing unit in the form of messages from a source external to said system;
• and
• an output interface for providing result information derived from selected ones of said messages to utilization means for external to said computing system,
• wherein said central computing unit transmits said messages to said independent units via said broadcast switch, said independent units compare said messages with said classifiers, and wherein selected result messages produced by selected ones of said independent processors are returned via said broadcast switch to said central computing unit.