Learning from Examples
Sanjeev R. Kulkarni, Gilbert H. Harman · Wiley series in probability and statistics · 2011
Knowledge of the distributions is replaced by a set of examples called training data. The problem of learning from examples is to use the training data to come up with a decision rule for classifying a new feature vector. Although the distributions are unknown, the performance of the decision rule learned is evaluated using the unknown distributions. A brute force approach to learning from examples is to use the training data to estimate the unknown distributions and then construct a Bayes decision rule using our estimates. This is impractical in most applications because of the difficulty of estimating densities in high dimensions with minimal data. Though methods will be discussed to avoid this problem to some extent, learning from examples is inherently a difficult problem, especially in high dimensions. This is often called the curse of dimensionality. Controlled Vocabulary Terms Bayes' theorem