A Necessary Condition for Learning from Positive Examples

作者:Haim Shvaytser

摘要

We present a simple combinatorial criterion for determining concept classes that cannot be learned in the sense of Valiant from a polynomial number of positive-only examples. The criterion is applied to several types of Boolean formulae in conjunctive and disjunctive normal form, to the majority function, to graphs with large connected components, and to a neural network with a single threshold unit. All are shown to be nonlearnable from positive-only examples.

论文关键词:Concept learning, learning from positive examples, nonlearnability, sample complexity, predictable concepts

论文评审过程:

论文官网地址:https://doi.org/10.1023/A:1022663809420