Generalization and generalizability measures

Abstract

In this paper, we define the generalization problem, summarize various approaches in generalization, identify the credit assignment problem, and present the problem and some solutions in measuring generalizability. We discuss anomalies in the ordering of hypotheses in a subdomain when performance is normalized and averaged, and show conditions under which anomalies can be eliminated. To generalize performance across subdomains, we present a measure called probability of win that measures the probability whether one hypothesis is better than another. Finally, we discuss some limitations in using probabilities of win and illustrate their application in finding new parameter values for TimberWolf, a package for VLSI cell placement and routing. Index Terms-Anomalies in generalization, credit assignment problem generalization, machine learning, subdomains, probability of win, VLSI cell placement and routing.

Video tracking system optimization using evolution strategies

“…On the contrary, random selected examples could produce disorientation in the search. Thus, the set of data that optimize the search of the suitable parameters is defined as the ideal trainer (Wah, 1999). In this problem, each trajectory or moving target could be considered as an example, and the ideal trainer is a set of trajectories that represent different situations of the surveillance problem that should be reflected in the final design.…”

A bayesian conceptualization of space for mobile robots

“…In the artificial intelligence (AI) community, the problem of generalization has been well addressed. The work [13] provides a good overview of different generalization strategies that exist and how they relate to each other. The approach presented in this work can be likened to a data driven approach which requires a set of positive / negative exemplars (or a "teacher") to learn from.…”