Delta learning system for using expert advice to revise the Hierarchies of damage to the experts diagnostic systems.

摘要

A computer system takes as input a HIERARCHY OF INITIAL ERRORS KB {SUB, 0} and a set of session transcripts ANNOTATED C = {{IP} {SUB, J}, {RH} {SUB, J}} AND It IS PROVIDING A sPECIFIED OPERATORS review set T = {{DS} {SUB, J}} where each {DS} {SUB, J} ES T SITUA A HIERARCHY OF ERRORS KB in a nest {DS} {SUB, J } (KB). This hierarchy ERROR REVISED {DS} {SUB, J} (KB) IS CALLED A NEIGHBOR OF KB and the set of all neighbors of KB, N (KB) = {{DS} {SUB, I} (KB )} IS THE VICINITY OF KB. The computer system uses T to scale from hierarchy KB INITIAL ERRORS {SUB, 0} HIERARCHIES THROUGH SUCCESSIVE SUB KB {1} ... {SUB KB, M}, with higher empirical accuracy successively C. THE FINAL RANK, KB SUB {M}, IS A OPTIMAL POSITION IN SPACE DEFINED BY OPERATORS OF REVISION. EVERY STEP, TO GO FROM ONE RANK OF ERRORS KB {SUB, I} TO YOUR NEIGHBOR KB {SUB, I + 1}, the present invention TO EVALUATE THE ACCURACY OF KB {SUB, I} ON C and the PRECISION EACH KB EN N (KB {SUB, I}). IF YOU ARE ANY KB is more accurate than KB {SUB, I}, this hierarchy KB {SUP *} BECOMES THE NEW PATRON, LABELING KB {SUB, I + 1}, and the review process of theory rEPEATS, SEEKING A NEIGHBOR OF THIS KB {SUB, I + 1} which more accurate than KB {SUB, I + 1} ON JOINT EXAMPLES C. OTHERWISE, if no KB is more accurate than KB {SUB , I}, PROCESS CLIMBING A return to this KB {SUB, I} and end.