Support Vector Machines

• Russell and Norvig say that the SVM framework is currently the most popular approach for off the shelf supervised learning.
• It's what you start with for an analytics problem.
• It uses optimally placed linear separators (based on the support vectors).
• When linear separators don't work, it uses the kernel trick to project to higher dimensions where they do work.
• It's a framework because you need to select the appropriate kernels.
• So, an analyst uses this framework to get a cheap lunch.