Mikhail V. Kiselev
The present paper describes the learning technique used in PolyAnalyst - the system of machine discovery and intelligent analysis of the experimental/observational datawhich has been created in the Computer Patient Monitoring Laboratoryat the National Research Center of Surgery. Po!yAnalyst Is a multi-purpose system designed to solve the following classes of problems: I) construction of a procedure realiT.ing the mapping from the set of desCriptions to the set of parameters given by the pairs description, parameter; 2) search for the interdependences between componenm of the description; 3) search for characteristic features of agiven set of descriptions. Here the description means a single experimental/observational data record and it is assumed that all the records in the same set of observations have the same structure. The peperis devoted mainly to P01yAnalyst’s application to the ~rpe 1 problems, This type includes classification, empirical law inference, choice of the best decision from a fixed set of possible decisions, and other tasks. To solve a problem PoIyAnalyst constructs and tests programs on a simple functional programming:language whose inputs are the descriptions and outputs are the corresponding parameter values. While searching forthe solutionPolyAnalyst combines full search, heuristicalsearch, and direct cons~'uction of the programs. Since the first version of PolyAnalyst was created it has solved a number of real problems from chemistry, medicine, geophysics, and agricultural science. One example is given in the paper - the prediction of the elasticity of the polyethylene samples from their infra-red spectrums.