Parametric vs. Non-Parametric Learning Algorithms

• There are lots of ways to subdivide learning algorithms, but one is between parametric and non-parametric algorithms.
• Parametric algorithms use the data to learn the parameters, and then they throw the data away.
• An MLP is an example of this. You use the data to learn the weights (the parameters).
• Once you've learned the weights, it's very efficient.
• A non-parametric algorithm keeps the data around.
• So, the Euclidean distance measurement I suggested in lab 16, or that is often used in case-based reasoning systems is non-parametric.
• The problem with a non-parametric algorithm is that it can really slow down when there is a lot of "training" data.
• Is a GA paramteric or non-parametric?
• How about conditional probability derived from input?