Machine Learning: A Theoretical Approach

Valiant's PAC learning model surely must rank as one of the most fruitful ideas introduced into computer science in the last decade. "PAC" stands for "probably approximately correct," which, despite sounding like a grammatical joke, is a compelling model. It permits a learning algorithm two parameters of slack: precision ε ("approximately correct") and confidence δ ("probably"). The beauty of the model is that it lends itself to analysis, amply demonstrated in this book.