Recently read Professor Lin Xuan Tian Taiwan University "machine learning" courses, to sum up the course to explain the linear models and kernel models, curriculum summarized the position of "machine learning technique" 6-4 (in the sixth week, the fourth section)
Screenshot of video
linear models (linear models of a summary map)
can see that we are familiar linear model in the field of machine learning from the figure.
Kernel linear models and model summary chart (linear / kernel models)
can be seen from the figure, in addition to previously common linear model, has joined the kernel model.
This chart summarizes the good contact linear model of machine learning and kernel model, through these connections, we may be able to create new models to better meet the actual situation.
1. In real life, the first row in FIG, PLA / pocket, linear SVR is not commonly used, because they do not fit well in the actual data. They are too simplistic, requiring data with good characteristics in order to achieve good results, and the data obtained in real life we often complex.
2. The third line in the figure, kernel ridge regression and kernel logistic regression is relatively less, we are solving these models, it is difficult to get the majority of β 0, and dense β make us spend a lot of time in the forecast effort, resulting in a large amount of calculation.
3. For the second row in FIG often used instead of the algorithm in the first row. The fourth line, often algorithm instead of using the third row.
So in real life, we should be according to our problems, priority to select the appropriate algorithm to deal with our problems, rather than to the algorithm we have learned to try. Take the time, do pre-selection tests and, more often than later improved results we are dealing with the problem would be better.
In this video thank Professor Lin Xuan Tian Taiwan University to explain.
