Read the relevant books first: there should be several books for reference, preferably from different angles and depths, such as systematic introduction, in-depth derivation and simple explanation. It should be used as a reference book in different learning processes.

Online tutorials: For example, Andrew Ng's blogs on various topics, find the corresponding tutorials in a targeted way. Relatively speaking, online resources such as courses of 5 1cto College are not equivalent to books, but also very useful sources of information.

Papers: For example, when you are studying CNN, you will search for a bunch of papers and concentrate on reading them for a while. You must read them with a clear question mark.

Open source resources: Many things are actually open source, used directly, understood while using, and even learned the source code.

Pay attention to this field: pay attention to relevant conference journals, and pay attention to the dynamics of Daniel (Hinton, Bengio, LeCun, etc. ), and add a bunch of Daniel in Weibo to see what they share every day.

Basic knowledge: line generation, statistics, probability, mathematical analysis; Basic concepts in information theory should be understood, such as knowing what relative entropy means and how to calculate it; Convex optimization and optimal estimation will be encountered in various problems, and learning it well can be of great help.

Machine learning is mainly to find the objective function and estimate the parameters. Although there are many ready-made tools, you will be blind if you are not familiar with the optimization problem.

Familiar with several models: neural network (SAE, RBM, CNN, etc. ), SVM, maximum entropy, CRF, random forest, GMM, etc. Understand the advantages and disadvantages of various models in different application scenarios, and select some exercises that you may use frequently in the future.