New Approaches to Two Tasks of Knowledge Discovery: Niche Finding (new) and Benchmarking (old)
1. NICHE FINDINGGiven a block of data (rows=individuals, columns=numeric/symbolic features), three common tasks are to (1) learn a classifier of future individuals, (2) cluster the individuals, and (3) find associations among the feature values. I will introduce a new task with a comparable applications potential: Concisely articulate what is unique about a specific individual. I will mention three pilot applications to world languages, U.S. univiersities, and U.S. Congressmen, and sketch others in functional genomics and baseball announcing.2.BENCHMARKINGGiven N alternative designs, if a benchmarking study reveals a clear winner, then (maybe) you are done: report the best. If the outcome is less clearcut, then one should extract knowledge about the comparative advantages of the designs. I will apply our recently-developed profiling methods to articulate concise knowledge such as "Design A is better than designs B,C,D on problems having property P. and better than E,F on designs having property Q". I will mention collaborative applications to numerical methods (design = algorithms), and chemical catalysis (designs = catalysts, benchmarks=reactions).