PRASAD@RED.RUTGERS.EDU (03/18/86)
Machine Learning Colloquium LEARNING ARGUMENTS OF INVARIANT FUNCTIONAL DESCRIPTIONS Mieczyslaw M. Kokar Northeastern University 360 Huntington Avenue Boston, MA 02115 11 AM, March 25, Tuesday #423, Hill Center The main subject of this presentation is discovery of concepts from observation. The focus is on a special kind of concepts - arguments of functional descriptions. The functions considered here are to be meaningful, i.e., computable functions expressed in terms of the operations defining the representation language in which the concepts are described. Such functions are invariant under transformations of the representation language into equivalent representations. It will be shown that the feature of invariance can be utilized in formulating and testing hypotheses about relevance of arguments of functional descriptions. The main point is that the arguments do not need to be changed to test the relevance. This is very important to the discovery process as the arguments to be discovered are not known, therefore, how could they be controlled? Simple examples of discovering concepts of physical parameters (arguments of physical laws) will be discussed. ** We need a host for the speaker. Please send me a message soon if you want to host him. -------