SHORT-TERM LEARNING IN CONFORMATIONAL-ANALYSIS

A method for learning short-term rules of conformational analysis is i ntroduced. The technique works by discovering problems during the buil ding of a conformation in Cartesian coordinate space, and builds an ab stract critic suitable for reasoning in abstract symbolic space. The m ethods not only afford speed increases ranging from 1.0- to 2.3-fold i n WIZARD (analysis completed in 100% to 43% of original run time), but can be modified to provide similar increases in other programs that u se ''template joining'' and distance geometry. These methods also prov ide the basis for a long-term learning project.