1. What are the contributions mentioned in the paper "Data-efficient information-theoretic test selection" ?
The authors describe and compare a basic approach consisting of averaging MI estimates conditioned on individual observations and another approach where it is possible to condition on all observations at once by making some conditional independence assumptions.. The authors present experimental results on public heart disease data and data from a controlled study in the area of breast cancer diagnosis.. 1 Information Maximization for Medical Test Selection Consider a collection of patient records containing data generally indicative of various clinical aspects associated to the patient, e. g., patient demographics, reported symptoms, results from laboratory or other tests, and patient disease/condition.. In this paper, the authors consider the set of variables V = { V1,..., VM } and they assume each variable to be discrete with a finite domain.. Finally, for the patient, let Y denote a variable in which the authors are ultimately interested, but that they have not observed, such as the occurrence of cancer.. Formally, the authors want to optimize X∗ = argmax j I ( Xj, Y |Z1 = z1,..., Zk = zk ), where the quantity I is the mutual information ( MI ) between their variable of interest Y and the ( test ) variables whose values they could potentially obtain Xj, conditioned on the fact that they already know Z1 = z1,.
read more
