anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Technion</strong> - Computer Science Department - Ph.D. Thesis PHD-2008-12 - 2008<br />
Chapter 1<br />
Introduction<br />
Machine <strong>learning</strong> techniques are gaining prevalence in the production of a wide<br />
range of <strong>classifiers</strong> <strong>for</strong> complex real-world applications with nonuni<strong>for</strong>m testing<br />
and misclassification costs. The increasing complexity of these applications poses<br />
a real challenge to resource management during <strong>learning</strong> and classification.<br />
Assume, <strong>for</strong> example, that a medical center has decided to use machine <strong>learning</strong><br />
techniques to build a diagnostic tool <strong>for</strong> heart disease. Figure 1.1 shows three<br />
possible trees. The first tree (upper-left) makes diagnoses using only the results of<br />
cardiac catheterization (heart cath). This tree is expected to be highly accurate.<br />
Nevertheless, the high costs and risks associated with the heart cath procedure<br />
make this decision tree impractical. The second tree (lower-left) dispenses with<br />
the need <strong>for</strong> cardiac catheterization and makes a diagnosis based on a single, simple,<br />
inexpensive test: whether or not the patient complains of chest pain. Such<br />
a tree would be highly accurate: most people do not experience chest pain and<br />
are indeed healthy. The tree, however, does not distinguish between the costs of<br />
different types of errors. While a false positive prediction might result in extra<br />
treatments, a false negative prediction might put a person’s life at risk. There<strong>for</strong>e,<br />
a third tree (right) is preferred, one that attempts to minimize test costs<br />
and misclassification costs simultaneously.<br />
As another example, consider a hardware manufacturer who uses a machine<strong>learning</strong><br />
based tool <strong>for</strong> assuring the quality of produced chips. In realtime, each<br />
chip in the pipeline is scanned and several features can be extracted from the<br />
image. The features vary in their computation time. The manufacturer trains<br />
the component using thousands of chips whose validity is known. Because the<br />
training is done offline, the manufacturer can provide the values of all possible<br />
features, regardless of their computation time. In realtime, however, the model<br />
must make a decision within 2 seconds. There<strong>for</strong>e, <strong>for</strong> each chip, the classifier<br />
may use features whose total computation time is at most 2 seconds. Unlike<br />
the medical center, the manufacturer is not interested in minimizing the total<br />
5