Author/Editor		Todorovski, Ljupčo
Title		Omejevanje prostora hipotez pri odkrivanju enačb
Translated title		Declarative bias in equation discovery
Type		monografija
Place		Ljubljana
Publisher		Univerza v Ljubljani, Fakulteta za računalništvo in informatiko
Publication year		1998
Volume		str. 59
Language		eng, slo
Abstract		In the areas of machine learning dealing with very expressive and large hypothesis spaces, declarative bias is used to define and restrict the search space for the target concept. Although equation discovery systems deal with potentially infinite hypothesis spaces, only few (if any) make use of declarative bias to specify the form of the discovered equations. The thesis presents a declarative bias formalism for equation discovery. We use context free grammars to specify the form of discovered equations. These grammars can make use of functions defined on the basis of domain specific knowledge, in addition to common mathematical oeprators. The formalism is implemented in the equation discovery system LAGRAMGE that heuristically searches the equation space defined by the context free grammar. LAGRAMGE is applied to several equation discovery problems in artificial and real world domains. Among others, the complex model of an inverted pendulum that consists of a cart and two poles was successfully reconstructed from numerically simulated data. LAGRAMGE was also used for automated modeling of phytoplankton growth in the Danish lake Glumsoe. The discovered model is suitable for accurate short and long term prediction of the phytoplankton concentration in the lake.
Descriptors		ARTIFICIAL INTELLIGENCE PHYTOPLANKTON ECOSYSTEM NUMERICAL ANALYSIS, COMPUTER-ASSISTED FRESH WATER DECISION TREES DENMARK