1.1 Standard Description of Learning Problems (P2)
1) Three features: Task of task; standards for measuring task; source of experience
1.2.1 Select training experience (P4)
1) Can training experience provide direct or indirect feedback for system decisions
2) Learning machine can control training sample sequences to a large extent
3) Training sample can express the instance distribution, and measure the performance of the final system by sample.
1.3 Some opinions and problems of machine learning (P10)
1) Machine learning problem is often attributed to search questions, which is to search for a very large hypothetical space to determine a better assumption that the best fitted data and learner have knowledge.
1.5 Summary (P12)
0) Machine learning is committed to establishing a computer program capable of improving processing performance based on empirical self
1) The machine learning algorithm has proven to be practical value in many fields. Such as: analysis of treatment results; from the image library in the image library; production control; adapt to changes in personal reading interest
2) A complete machine learning requires: a clear defined task; performance metrics are used; the source of training experience
3) The algorithm process contains many options: the type of training experience; the objective function of learning; the representation of the target function; the algorithm of learning the target function from training
4) The process of learning the process is the search process, search for possible hypotheses, making the assumption that the hypothesis is most in line with existing training samples and other pre-constraints and knowledge
(You should have a good memory from the beginning and master all the concepts, this book is particularly very much, and it is not good to understand!)
