TY - GEN
T1 - Super-boxes - an accountable advisor system
AU - Russell, D. W.
PY - 1993
Y1 - 1993
N2 - The paper describes a radical modification to the well-known BOXES methodology. Learning with the BOXES paradigm is slow and capricious, and it is proposed that the insertion of voting logic will accelerate and improve the process, particularly in the case of poorly defined systems. Advisor states are selected by a modified BOXES state quantization algorithm, rated for positional strength and made accountable, statistically speaking, for their contribution to the performance of the system, by a new post-processing algorithm. In the spirit of the conference theme, a reference is made to the possible application of the method to the control of a municipal trash-to-steam plant.
AB - The paper describes a radical modification to the well-known BOXES methodology. Learning with the BOXES paradigm is slow and capricious, and it is proposed that the insertion of voting logic will accelerate and improve the process, particularly in the case of poorly defined systems. Advisor states are selected by a modified BOXES state quantization algorithm, rated for positional strength and made accountable, statistically speaking, for their contribution to the performance of the system, by a new post-processing algorithm. In the spirit of the conference theme, a reference is made to the possible application of the method to the control of a municipal trash-to-steam plant.
UR - http://www.scopus.com/inward/record.url?scp=0027210308&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0027210308&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:0027210308
SN - 1851668381
T3 - Applications of Artificial Intelligence in Engineering
SP - 477
EP - 489
BT - Design, Methods and Techniques
PB - Publ by Computational Mechanics Publ
ER -