Publication Date:
2019
Description:
〈h3〉Abstract〈/h3〉
〈p〉We introduce the speculate-correct method to derive error bounds for local classifiers. Using it, we show that 〈em〉k〈/em〉-nearest neighbor classifiers, in spite of their famously fractured decision boundaries, have exponential error bounds with 〈span〉
〈span〉\(\hbox {O} \left( \sqrt{(k + \ln n)/n} \right) \)〈/span〉
〈/span〉 range around an estimate of generalization error for 〈em〉n〈/em〉 in-sample examples.
〈/p〉
Print ISSN:
0885-6125
Electronic ISSN:
1573-0565
Topics:
Computer Science