Learning Compatible Models

 

Learning Compatible Models

Sage 4101

A modern way of approaching large data sets in various fields, ranging from genomics to financial databases, stays in abandoning the goal of reaching the truth in favor of gaining a suitable organization of the knowledge we may grasp from the data. We encompass it into a model, where suitability stands for compatibility with the data themselves. While the truth is unique by definition, models may be many and we manage this variety by smearing probabilities on them. With this key we revisit some families of learning instances, realizing how and why it is possible to get rid of them, also when we are far from the standard frameworks (made up of linearity, gaussian distributions and independence) where classical mathematical tools have chance to succeed.

Add to calendar
Share|