Machine Learning

Introduction Human beings can extend their knowledge to adapt to changing circumstances. To do this, they need to "learn". Learning can be easily defined as learning knowledge and skills by learning, experience, or being taught. Although learning is easy for most people, obtaining new knowledge and skills from the data is too difficult for the machine. Furthermore, the intelligence level of a machine is directly related to the ability to learn. Machine learning research is trying to cope with this complicated task.

Machine learning can only be as good as the data used to train it. The phrase "garbage in, garbage out" is before machine learning, but it correctly represents the main limitation of machine learning. In machine learning, only patterns that exist in training data can be found. Supervised machine learning tasks such as classification require a powerful, superior, feature rich training data set. Machine learning is effective only when training data is representative. Machine learning only works on data generated by the same distribution that generates training data, as the Fund's prospectus warns that "past achievements can not guarantee future results" You need to be warned that you can guarantee to do so. Be careful of the deviation between training data and production data. Also, we often retrain the model so that the model does not become stale.

Machine learning is about data and algorithms, but it is mainly data. There are many exciting things about machine learning algorithms, especially with respect to the progress of deep learning. However, data is an important element enabling machine learning. You can do machine learning without complicated algorithms, but there is no good data. Without a lot of data, you should stick to a simple model. Machine learning lets you learn models from data patterns and explores the possible model space defined by parameters. If the parameter space is too large, use training data too much to train a model that can not be exceeded. A more detailed explanation requires more mathematics, but usually it is necessary to simplify the model as much as possible.