Sparse Fuzzy Modeling - Nikhil R Pal - WCCI 2016

"Nature is pleased with simplicity. And nature is no dummy” -Isaac Newton

“I ask many questions, and when the answer is simple, then God is answering.” -Albert Einstein

These quotes by the two greatest scientists of all time suggest a very important point: “truth” is simple, the principle of parsimony. Taking inspiration from this, Nikhil R Pal notes that when scientists and technologists make a model to explain some natural phenomena, they should always look for a simple model that works. In other words, “Make everything as simple as possible, but not simpler.” (Also Albert Einstein.) 

Sparse modeling is a particular manifestation of this principle of parsimony; it is one of the ways to realize “simple models”. The literature on sparse modeling in statistical learning is quite rich, but in fuzzy modeling, although it is sometimes used as a good principle of design, it is not that active an area of research. 

In this talk, Pal briefly discusses a few illustrative approaches to classical sparse modeling in statistical learning and then talks about some of the attempts in fuzzy modeling. In this context, he considers three types of problems: classification, clustering and regression. Finally, he concludes with some results of his own investigation in sparse fuzzy modeling.